Deep Reinforcement Learning

Existing Vision-Language Models (VLMs) exhibits a critical bottleneck in robust spatial reasoning. Recent reinforcement learning (RL) methods aim to close this gap with verifiable outcomes, yet they suffer from poor credit assignment across intermediate reasoning steps. Concurrently, structured reasoning approaches overlook the critical depth perception necessary for comprehensive 3D understanding. To address these challenges, we propose SCOUT (Structured Chain-Of-Thought Utilizing Process-Supervised RL Training). Specifically, we design a structured Chain-of-Thought (CoT) framework that explicitly models 3D environmental perception to ensure robust spatial understanding and reasoning. Furthermore, we introduce a novel RL algorithm featuring multi-objective process rewards and a tailored advantage estimation method, facilitating fine-grained credit assignment across distinct segments of the reasoning trajectory. To support our framework, we develop SCOUT-24k, a structured spatial reasoning CoT dataset synthesized through a customized pipeline. Extensive evaluations demonstrate that SCOUT-3B improves upon baseline models by 16.85% and 6.3% on general spatial benchmarks and complex spatial reasoning tasks respectively. Notably, our larger SCOUT-7B even outperforms GPT-4o by a margin of 4.28%. Moreover, despite being trained exclusively on single image, SCOUT-7B exhibits robust out-of-domain generalization to multi-image and video scenarios. These empirical results render SCOUT as a critical step towards next generation of spatially-aware VLMs.

Integrating locomotion and manipulation is essential for robot autonomy, but scaling standard Reinforcement Learning (RL) to complex tasks is severely bottlenecked by the slow, manual process of dense reward shaping. To bypass this limitation, we leverage Sample-based Model Predictive Control (SMPC) entirely in simulation as an automated, rapidly tunable expert to generate massive offline datasets. Because this data solves the fundamental exploration problem, we can train an off-policy RL agent using purely sparse task rewards, drastically reducing the time required to learn new skills and eliminating the need for manual tuning. Integrating this high-level agent with a low-level dynamic stability controller yields more optimal behaviors that strictly align with true task objectives, ultimately allowing the learned policies to surpass the original optimal control teacher. We validate the robustness of this sim-to-real framework by successfully deploying complex loco-manipulation skills across different morphologies, including an arm-equipped Spot quadruped and a G1 humanoid.

Large language model agents increasingly rely on long-horizon reasoning to solve complex tasks involving planning, tool use, and memory. A critical capability in such settings is reflection: assessing trajectory progress, identifying missing evidence and unreliable intermediate states, and deciding whether to continue, revise, or abandon the current branch. Learning effective reflection, however, is challenging because reflection is performed locally within the current branch, whereas its utility can only be determined by its contribution to the final trajectory outcome. This local-global mismatch makes outcome-based reinforcement learning provide only local, sparse and delayed supervision for reflective decisions. To solve these, we propose LoongReflect, a training framework that formulates reflection as a memory-control policy. The agent operates over a reversible trajectory tree using explicit reflect and backtrack actions. Reflection consolidates verified facts, missing evidence, and branch-specific risks into working memory, while backtracking removes an unreliable branch from the active context and preserves a concise corrective lesson. To learn this policy, LoongReflect combines two complementary signals through a look-ahead, extragradient-style coordination mechanism. A fast channel distills globally informed reflective behavior from a privileged teacher, with supervision restricted to reflection and backtracking tokens. A slow channel optimizes complete trajectories using outcome-based GRPO, aligning local control decisions with final task success. Experiments on multi-hop retrieval-augmented generation and mathematical reasoning benchmarks demonstrate consistent improvements over outcome-only reinforcement learning and self-distillation baselines.

The reliability of Large Language Models (LLMs) for API calling degrades in multilingual settings. A common failure occurs when a model selects the correct tool but generates argument values in an inconsistent language, which we term Argument Language Mismatch (ALM). Although semantically correct, such outputs are operationally invalid and not captured by standard API-calling metrics. We revisit post-training strategies for mitigating ALM and find that, in our benchmark, supervised fine-tuning (SFT) provides a strong baseline, substantially improving argument language consistency and end-to-end function call accuracy. Under consistent model selection, SFT achieves performance comparable to, and sometimes exceeding more complex reinforcement learning (RL) approaches. We further examine whether RL with structured, argument-aware rewards offers additional benefits. While methods such as Group Relative Policy Optimization (GRPO) can improve language consistency and better preserve general reasoning ability, these gains are incremental and most pronounced in generalization and multi-objective trade-offs. Overall, our results suggest that much of the performance in multilingual API grounding can be achieved through careful supervised training, with RL providing targeted rather than fundamental improvements.

Offline reinforcement learning (RL) offers considerable promise for optimizing ICU treatment decisions, yet standard evaluation metrics Mean Squared Error (MSE) and Fitted Q-Evaluation (FQE) assess only behavioral imitation and cannot detect Toxic Mimicry, a failure mode in which agents replicate harmful patterns such as treatment withdrawal during comfort-care transitions. Using the MIMIC-III database, we propose the Counterfactual Clinical Audit (CCA) framework, which stress-tests RL agents through physiological perturbations anchored in Surviving Sepsis Campaign (SSC) guidelines. We audit a Medical Decision Transformer (MedDT) and a Historical Causal Transformer (HCT-RL), the latter employing Causal Action Shielding, propensity-based importance weighting, and Conservative Q-Learning. CCA reveals that MedDT paradoxically reduces vasopressor dosage as lactate escalates, contradicting resuscitation guidelines, while HCT-RL maintains physiologically consistent responses. These findings expose a systemic misalignment between statistical fit and clinical safety, supporting counterfactual audits as a necessary evaluation standard for medical RL.

GUI Visual Grounding is a fundamental capability for GUI agents. Existing models typically freeze their parameters after deployment, limiting their ability to adapt to unseen interfaces. Although recent methods attempt to adapt models via test-time reinforcement learning, they cannot reflect upon failed exploration. To overcome this, we propose a Test-Time Self-Evolving framework that enables models to improve after deployment without human-annotated ground truth. It constructs a closed-loop of Exploration, Evaluation, Reflection, and Internalization. Specifically, the agent first explores unseen interfaces by predicting grounding coordinates for given instructions. To evaluate these explorations, we introduce an MLLM-based Reflector to assess the generated results and provide the corresponding reasoning reflections. To internalize reflection knowledge into the model weights, we propose Reflection-Guided On-Policy Self-Distillation, which translates high-level reasoning into dense token-level supervision via a conditioned self-teacher. Furthermore, we design a Contrastive Calibration method to prevent incorrect auto-regressive prefixes from corrupting the supervisory signals during failed explorations. Extensive experiments across six benchmarks demonstrate our framework's effectiveness, achieving an average accuracy improvement of 7.4% over the base model. To the best of our knowledge, this is the first work to successfully exploit on-policy self-distillation for test-time adaptation in GUI visual grounding. By filling the gap in post-deployment adaptation, our framework completes the self-evolving capability of GUI agents. The code will be released.

Agent skills represent a standardized format for packaging procedural knowledge and domain expertise, serving within agent harness systems as an essential mechanism to continually constrain a language model's behavior space for repeatable, high-quality task execution. However, because strong closed-source models entail high inference costs, current popular agent harnesses, such as Codex and OpenClaw, remain prohibitively expensive when deploying these skills to accomplish real-world tasks. The rapid capability enhancement of open-source models deployable on consumer-grade GPUs presents a compelling opportunity to drastically reduce these costs by leveraging skill-based behavioral constraints. Nevertheless, automatically generating effective skills tailored specifically for such compact models remains a significant practical challenge. To address this, we propose SKILLER, a natural-language-driven reinforcement learning framework designed to automatically generate executor-specific skills for small models, which employs a strong model as the actor and critic, treats the small-model agent system as the environment, and propagates all reinforcement learning signals entirely via natural language. Extensive experimental evaluations across five relevant benchmarks using Qwen3.5-9B and Qwen3.5-4B demonstrate that SKILLER outperforms three open-source and one closed-source skill generation or evolution methods, achieving absolute gains ranging from 4.3 to 20.4 percentage points for the 9B model and 1.8 to 13.3 points for the 4B model, while remarkably matching the performance of strong closed-source models on single-skill tasks in SkillsBench. The project is available at https://github.com/DANG-ai/SKILLER.

Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones. We propose SafeCap, a reinforcement-learning framework that aligns LVLMs through learned self-captioning. SafeCap trains a policy model to first generate a safety-relevant image caption and then produce a final answer; the caption is further optimized by whether it enables a frozen LLM to reach a safety-aligned decision. This caption-mediated objective encourages the policy to expose visual cues relevant to safe response generation rather than relying solely on direct refusal supervision. Across five multimodal safety benchmarks and six vision-utility benchmarks, SafeCap substantially improves aggregate safety performance under its intended DirectCap protocol, with gains of 3.7-19.0 points in safety average across four model settings while maintaining comparable or improved vision utility. Under controlled comparisons on matched backbones and data, SafeCap outperforms safety SFT, DPO, and SafeGRPO, demonstrating the effectiveness of caption-mediated reinforcement learning for multimodal safety alignment.

Long-horizon tool-using agents must reason over user goals, domain policies, tool calls, simulator state, and delayed verifiable rewards. Reinforcement learning (RL) is a natural fit for this setting, but multi-turn on-policy rollouts create long contexts, while model-specific attention layers may require custom masks and learned sink normalization. We present SINKFLEX-RL, a modular training system for RL in dual-control tool-use environments. The system combines a Gymnasium-compatible environment wrapper, a VERL-style rollout dataflow, group-relative policy optimization without a separate value model, and a sink-aware FlexAttention path designed to preserve model-specific sink scaling under causal and sliding-window masks. In a preliminary Tau2Bench retail run, validation reward (mean@1) rises from 0.25 early in training to $0.44$ later in the observed training window, while training-score and trajectory-reward proxies also trend upward. In a fixed-configuration memory benchmark, the optimized attention path reduces peak VRAM from 28.06GB to 22.52GB at 4096 tokens, a $19.7\%$ reduction, and runs the measured 8192-token configuration using $25.53$~GB where the eager baseline runs out of memory. These results illustrate the value of integrating environment interfaces, RL dataflow, and attention-kernel design for memory-feasible long-horizon agent training.

Safe reinforcement learning maximizes reward subject to safety constraints. For Constrained Markov Decision Processes, the linear-programming view over occupancy measures implies that whenever the constraint is active at optimality, the optimal policy lies exactly on the constraint boundary, yet standard gradient-based methods do not exploit this structure and often settle in the feasible interior. We introduce Boundary-Seeking Policy Gradient (BSPG), a first-order method whose update combines a tangential component that improves reward while preserving cost to first order with a signed, residual-driven normal component that regulates the policy toward the active boundary from either side; the combined direction admits an algebraic Lagrangian form with an induced coefficient and no learned dual variable. Under exact gradients and stated regularity conditions, the constraint residual converges to zero from either side with a finite-horizon $O(1/\sqrt{T})$ bound, the tangential component is a reward-ascent direction on the boundary, and any convergent parameter sequence is stationary on the active constraint set, satisfying the KKT conditions when the limit is also a local maximizer over the feasible set. This complements existing analyses, which certify feasibility but do not characterize the constraint value at convergence. On a standard Safety-Gymnasium navigation task, BSPG attains higher reward while tracking the boundary more tightly than the compared baselines.

Exploration has been a focus of reinforcement learning research for a long time. Recently, there has been growing evidence that it is also an important ingredient in LLM reinforcement learning recipes that can significantly impact downstream performance. Many existing methods control exploration in the action-space, for example, using temperature scaling. However, these methods cannot reorder tokens but only influence the variance in the output distribution. This limits exploration and can lead to divergence or stalled training. Here, we investigate parameter-space exploration, where rollouts are generated by sampling different policies from a posterior that may each explore different rollouts. Sampling less or more diverse policies is then a complementary control lever over exploration. We introduce a family of methods called Perturbed Parameter Policy Optimization (3PO) which use different sampling strategies and different rollout grouping for reward estimation. Experiments on OLMo-3-1025-7B and Qwen2.5-Math-7B across mathematical reasoning and code generation tasks show that these approaches consistently improve average downstream performance over standard GRPO at a near-identical FLOPs cost. Moreover, using multiple parameter samples consistently produces fewer zero-advantage groups and malformed or incorrect rollouts during training than GRPO and action-space baselines. Overall, our work presents evidence that parameter-space exploration can improve reinforcement learning for LLMs.

Time-aware dense audio captioning (TDAC) aims to generate multiple fine-grained attributes (dense) of the audio with precise time boundaries (time-aware). Existing methods struggle to achieve these two goals and mainly rely on supervised fine-tuning, yielding sub-optimal performance. While reinforcement learning (RL) shows promise, applying it to TDAC faces two main challenges: (1) existing rewards are too coarse to supervise multi-event, multi-attribute, and multi-relation descriptions in a fine-grained manner; and (2) temporal supervision is difficult for free-form captions, where flexible event-time expressions make reliable event-time correspondence challenging. To address these challenges, we propose AudioMap, a novel RL-based TDAC framework, which shifts to a unified cloze-and-choice reward paradigm. Specifically, we introduce the Evidence Sufficiency Reward (ESR) with an asymmetric hierarchical scoring mechanism to promote fine-grained accuracy and descriptive richness across diverse acoustic dimensions. Furthermore, we design the Event-Conditioned Temporal Reward (ECTR) to structurally bind timestamps to event semantics via temporal IoU, accompanied by a dual-curriculum learning strategy to facilitate the training process. Finally, to support this task, we construct the first time-aware fine-grained audio captioning dataset, AudioMapCap-44K, which contains 44K carefully annotated captions. Extensive experiments across diverse benchmarks show that AudioMap achieves state-of-the-art (SOTA) performance among open-source models and delivers competitive or superior results relative to proprietary models. Project page and release updates are available at https://github.com/ryysayhi/AudioMap.

Natural language user preferences provide an interpretable interface for LLM personalization. However, universal preference summaries often contain information irrelevant to a particular downstream task. Directly supplying the full preference summary therefore wastes context capacity and introduces cross-task distraction, while manually designing task-specific preference views is difficult to scale. In this work, we study \emph{task-specific preference adaptation}: given a universal user preference summary and a downstream task, derive a task-conditioned representation that preserves sufficient decision-relevant evidence while removing redundant context. To this end, we propose \textsc{AlignXada}, a training-free meta-learning framework that induces reusable textual refinement policies for adapting universal preference summaries to task-specific ones. The refinement policy is iteratively optimized by a meta learner through verbal reinforcement learning. Across 13 tasks and three downstream models (39 task--model cells), \textsc{AlignXada} achieves an average gain of 3.82 points, improving 33 cells while retaining only 22.8% of the original profile tokens and outperforming RAG in 36 cells. An extended faithfulness analysis further shows that the refined profiles remain largely grounded in the source preferences while preserving task-relevant personalization signals, suggesting that profile-side adaptation serves as a practical complement to universal memory construction for lifelong personalized agents.

Efficient text-to-image generation requires both reinforcement-learning (RL)-based reward alignment and few-step distillation, yet these procedures are typically performed sequentially, increasing training cost and risking the loss of reward gains during compression. We instead take an RL-native perspective: diffusion RL already generates reward-scored finite-step trajectories, whose intermediate states provide a natural source of distillation supervision rather than a disposable byproduct of sampling. Based on this insight, we propose REST (Reward-Enhanced Scored-Trajectory Distillation), a single-stage RL-distillation co-training framework that attaches a decoupled student to an arbitrary RL teacher. The student learns segment-wise from the teacher's evolving rollout trajectories while leaving the original teacher optimization unchanged. To prevent uniform imitation from preserving undesirable low-reward behaviors, we further introduce Advantage-Modulated Distillation (AMD), which transforms rollout advantages into signed weights over a base distillation loss. AMD strengthens supervision from preferred trajectories and mildly repels the student from low-reward ones. The resulting framework is lightweight and plug-and-play, requires no extra image rollouts, no separate distillation dataset, and no adversarial training. Experiments on compositional generation, visual text rendering, and human-preference alignment show that REST enables few-step CFG-free inference that matches or surpasses its 40-step RL teacher, with an overall additional training cost below 25% over pure RL. REST improves DrawBench PickScore over RTDMD by 0.82 while requiring only one-fifth of the training iterations.

While learned robotic policies hold promise for advancing generalizable manipulation, their practical deployment is often hindered by suboptimal execution speeds. Imitation learning policies are inherently limited by hardware constraints and the speed of the operator during data collection. In addition, there are no established methods for accelerating policies learned via imitation, and the empirical relationship between execution speed and task success remains underexplored. To address these issues, we introduce SpeedTuning, a reinforcement learning framework specifically designed to enhance the speed of manipulation policies. SpeedTuning learns to predict the optimal execution speed for actions, thereby complementing a base policy without necessitating additional data collection. We provide empirical evidence that SpeedTuning achieves substantial improvements in execution speed, exceeding 2.4x speed-up, while preserving an adequate success rate compared to both the original task policy and straightforward speed-up methods such as linear interpolation at a fixed speed. We evaluate our approach across a diverse set of dynamic and precise tasks, including pouring, throwing, and picking, demonstrating its effectiveness and robustness in enhancing real-world robotic manipulation. Videos and code are available at https://daivdyuan.github.io/speed-tuning/

Aligning Large Language Models (LLMs) for open-ended tasks is challenging because responses must satisfy multidimensional criteria without following a single correct generation trajectory. Existing rubric-based reinforcement learning (RL) methods compress fine-grained criterion-level feedback into scalar rewards, making persistent capability gaps difficult to target under limited on-policy exploration. We propose $\textbf{RISE-RL}$ (Rubric-Informed Selective Exploration), which uses repeatedly missed rubric criteria to elicit privileged trajectories that are difficult to discover through unguided exploration alone. RISE-RL retains only trajectories whose complete-rubric reward exceeds the mean reward of natural rollouts, and then re-evaluates them under the original prompt to emphasize behaviors that remain weakly supported by the natural policy. The resulting guidance signal is optimized through a separate auxiliary objective and removed once its additional benefit diminishes. Experiments with 4B and 14B models across writing, chat, health, and science show that RISE-RL achieves the highest mean score on every evaluated benchmark under guidance-free evaluation. Compared with standard Rubric-RL, it improves the average score by 1.3 points at the 4B scale and $\textbf{3.3 points at the 14B scale}$, including a $\textbf{6.0-point}$ gain on CreativeWriting-V3. It also improves creative-writing diversity and yields gains on objectively scored medical and scientific benchmarks. These results indicate that selective internalization through reward filtering and policy support shaping is effective for open-ended reinforcement learning.

With the continuous improvement of computational capabilities, end-to-end reinforcement learning has been rapidly developed for remotely operated vehicles control. Nevertheless, existing end-to-end reinforcement-learningbased methods still face challenges in achieving optimal control under oceancurrent disturbances. In particular, there remains a lack of a unified control framework that can simultaneously achieve low steady-state tracking error, rapid transient response, energy-efficient operation, and smooth controlforce outputs under disturbances. To address the issue, this paper proposes the thrust smoothness rapid current adaptation proximal policy optimization (TSRCA-PPO) method which learns a near-optimal strategy by a twostage distillation learning framework. The core innovations of this work lie in the reward-function design and the privileged multi-encoder architecture. Ablation studies validate the effectiveness of each module. Simulation results demonstrate that the proposed TSRCA-PPO method consistently outperforms the conventional cascaded P-PID controller across all evaluation metrics. Specifically, TSRCA-PPO reduces the steady-state position error, steady-state attitude error, settling time, energy index, and thrustsmoothness index to 42.7%, 76.5%, 10.6%, 93.5%, and 15.9% of the corresponding P-PID values, respectively.

Video anomaly detection (VAD) is a critical yet challenging task due to the complex and diverse nature of real-world scenarios. Traditional deep learning approaches are fundamentally limited by poor generalization across diverse scenarios. While multimodal agents offer a promising tool-learning paradigm for VAD, current systems relying on supervised fine-tuning struggle with complex orchestration, and standard reinforcement learning often causes premature termination due to coarse-grained outcome rewards. To address these challenges, we propose VTO, a process-supervised reinforcement learning framework. Moving beyond static tool usage, VTO enables the agent to dynamically explore and interact with the environment. Specifically, we introduce a foundation model-driven cognitive evaluator to provide context-aware semantic feedback, which is seamlessly integrated into a Process-Supervised Cognitive Alignment that delivers fine-grained, step-wise supervision. By explicitly penalizing logical truncation and rewarding complete causal chains, the agent optimizes its multi-step reasoning policy for interrelated tool orchestration. To support our proposed framework, we meticulously crafted VAD-Tool, a hierarchical visual tool set comprising 12 specialized vision tools spanning from entity tracking to high-stakes hazard detection, and established the corresponding benchmark for rigorous multi-step reasoning evaluation. Extensive experiments on VAD-Tool demonstrate that VTO significantly outperforms baselines, achieving up to a 10.2% absolute accuracy improvement in tool scheduling. Code and data are available at https://github.com/MICLAB-BUPT/VTO.

Recent advancements in deep reinforcement learning have increasingly favored simplified, highly parallelized paradigms. Notably, the Parallelized Q-Network (PQN) algorithm enables off-policy value learning without relying on experience replay buffers or target networks. However, the representational capacity and computational efficiency of visual encoders operating in these buffer-free settings remain comparatively underexplored. In this work, we systematically investigate the architectural design space of Convolutional Neural Networks within PQN. We evaluate eight distinct CNN topologies while explicitly characterizing their parameter and computational requirements. We further study the effect of multiplicative representation learning and advanced value estimation by integrating the Hadamax encoding paradigm with categorical, ensemble, and dueling value heads. Extensive experiments on Atari-57 show that our final composite architecture, Aftab, achieves an Interquartile Mean (IQM) Human-Normalized Score of 6.592, compared with 2.715 for the standard PQN baseline, together with a 0.86 Probability of Improvement over PQN. We additionally evaluate Aftab on Procgen-Hard to assess performance under procedurally varying visual environments. Aftab achieves a normalized learning-curve Area Under the Curve (nAUC) of 0.541 compared with 0.216 for PQN. Overall, the results demonstrate that carefully designed encoder topology, multiplicative feature interactions, and advanced value-estimation heads can substantially improve performance within a parallelized, replay-free Q-learning framework while preserving its memory-efficient training paradigm. The complete Aftab framework, including model definitions, training configurations, reproducibility settings, and raw experimental logs, is open-sourced at https://github.com/tahashieenavaz/aftab

When a reinforcement learning agent cannot observe the full state, we usually blame its policies: it cannot see enough to represent a good one. We show that in a solvable case the bigger problem lies elsewhere. Even when a good policy is available and the agent's value function is expressive enough to describe it exactly, learning still ends up somewhere far worse. We study a partially observed linear-quadratic problem in which a standard actor-critic learner can be solved in closed form. At our default setting the best policy the agent can represent is already close to optimal, costing 10.4% more than the ideal controller that observes everything. Learning does not find it. The algorithm instead comes to rest at a policy that is 35% worse than the best one available to it, and we can say exactly where and why. The cause is a bias in what the critic learns rather than a limit on what the actor can express. Because the agent cannot attribute what it sees to the part of the state it cannot observe, the critic misreads that unexplained variation as sharp curvature in its own value estimates, and the actor follows that error away from the optimum. We derive closed-form expressions for the resulting policy, for its cost, and for the one design choice that removes the problem, which is how far the learner looks ahead before trusting its own value estimates. Deep reinforcement learning experiments follow these predictions closely. Notably, giving the agent memory of past observations does not help, while changing how far it looks ahead does.

Recent advances in deep reinforcement learning (RL) have shown that improving neural network architectures can yield substantial gains in sample efficiency and asymptotic performance without altering the underlying algorithms. In contrast, work on multi-objective reinforcement learning (MORL), which aims to discover a set of policies that balance trade-offs among conflicting objectives, has predominantly focused on algorithmic innovations, leaving the area of architectures underexplored. While the optimal policies and value functions can differ significantly depending on the trade-offs, MORL algorithms commonly represent them with simple feedforward networks conditioned on the trade-off. This raises the question of whether the performance of the algorithms could be improved with more expressive function approximators. In this paper, we integrate recent advances in neural network design: (i) observation and feature normalization, (ii) weight normalization, and (iii) modeling of distributional returns with an entropy-regularized MORL algorithm. The empirical results across standard continuous control benchmarks demonstrate that these changes substantially improve the quality of the produced solution sets without requiring major changes to the underlying algorithm.

Reinforcement learning policies are difficult to inspect, but interpreting them is a prerequisite for trustworthiness. Converting a trained policy into explicit decision-tree rules improves transparency and the resulting artifacts often remain too complex for human understanding. We present a pruning process that simplifies such rule-based policies while preserving task performance and making edits to the policy auditable. The process defines a small set of structural and usage-aware operators and evaluates candidate edits by re-executing the policy to measure return and interpretability proxies. This exposes an transformation process from complex to compact policy structures. We investigate this approach on classic control and MuJoCo benchmarks, where pruning traces reveal consistent interpretability improvements while maintaining high performance.

Reinforcement learning systems are significantly more complex than other machine learning paradigms due to inherent properties, causing RL system design to jointly account for many tightly coupled factors. Despite advances in individual algorithmic components, their functional interdependencies remain underexplored: do they exhibit mutual synergy or counterproductive interference? To bridge this gap, we conduct a systematic investigation and find that the efficacy of different components exhibits significant task-dependency, and naively stacking state-of-the-art techniques does not necessarily yield performance gains; instead, it often triggers emergent challenges, such as compounded non-stationarity. Building upon these findings, we distill a suite of actionable insights into the principled coordination of these components. Guided by these insights, we propose ROSER, an RL framework that coordinates three critical dimensions: Model-based Representation, Optimization Stability, and Experience Replay. Across diverse continuous-control benchmarks, ROSER consistently outperforms vanilla baselines and achieves 17.60% gains over naive stack. Our findings underscore the necessity of a holistic perspective in RL system design and paves the way for developing sample-efficient agents.

As an important component of the supply chain industry, transportation has experienced rapid development in the past decade with the assistance of digital platforms and intelligent algorithms. Within the field of transportation research, Vehicle Routing Problem (VRP) has remained a persistent and enduring challenge. In the realm of management science, experts, and scholars from both the industrial and academic sectors have continuously explored optimization models and algorithms to effectively address routing problems, from the classical Traveling Salesman Problem to the more general Vehicle Routing Problem. These models and algorithms are applied in real-world industrial scenarios to achieve cost optimization and reduce carbon footprints. However, due to the complexity of real-world problems, numerous specific constraints are often added, and challenges such as information opacity, uncertainty, and irrational human behavior may arise. Therefore, deploying and optimizing mathematical models for VRP in practical scenarios while maintaining optimal results poses numerous challenges. This paper discusses and provides solutions for three different logistic use cases involving external truck network design. Through these industrial case study, the paper introduces how deep reinforcement learning-based vehicle routing optimization has been implemented. As a result, it can be observed that the routes optimized by reinforcement learning agent have over 10% total cost compared to baseline results. Furthermore, the paper proposes that in future research, DRL algorithms for vehicle routing problems could be generalized into more variations of VRP.

World models for visual control typically learn compact latent states by reconstructing observations, implicitly encouraging representations to preserve information across the entire visual input. However, task-relevant content often occupies only a small fraction of the observation, while background clutter and distractors consume valuable representational capacity. This mismatch between visual reconstruction and control objectives biases latent representations to model task-irrelevant visual content, diluting learning signals for control-relevant features and severely degrading downstream performance under visual distractions. We introduce TaskSense, a task-centric world modeling framework that enforces task relevance before latent encoding through a differentiable stochastic spatial attention mechanism conditioned on the previous latent state. To steer attention toward control-relevant regions, we augment training with an auxiliary inverse-dynamics objective. Rather than reconstructing the full observation, the world model reconstructs only the attended regions, encouraging latent representations to preserve task-relevant information while discarding irrelevant visual content. The decoder is further conditioned on the sampled attention map, enabling consistent reconstruction despite stochastic attention. Compared with the DreamerV3 baseline, TaskSense maintains competitive performance on the DeepMind Control Suite while consistently outperforming DreamerV3 on the Distracting Control Suite, demonstrating substantially improved robustness to visual distractions. Qualitative analysis further confirms that the learned attention, guided by inverse-dynamics supervision, consistently localizes control-relevant regions while suppressing irrelevant visual content.

In simulation-in-the-loop decision-making systems, reinforcement learning (RL) inference is often constrained by simulator-side execution overhead, where workloads are highly dynamic and sensitive to runtime thread configurations. Existing multithreaded strategies struggle to match thread resources before or during execution, causing resource contention, scheduling overhead, and reduced throughput. Through empirical analysis, we identify the ratio of task execution time to scheduling time as the key factor determining the optimal thread count. Building on this insight, we propose AutoThread, a hybrid adaptive thread-tuning method for mitigating simulation bottlenecks in RL inference. AutoThread employs a Physics-Informed Neural Operator (PINO) as a thread-count predictor and incorporates a finite-source M/M/1 queueing model to constrain and guide prediction, enabling fast and accurate estimation under dynamic workloads. It further performs load-aware online fine-tuning to compensate for prediction errors and refine resource allocation. Experiments show that AutoThread improves average speedup by 18.4% over static strategies, achieves average throughput of 1.7x and 1.8x that of XGBoost and Reinforcer, respectively, and reduces execution time by up to 83.8% compared with state-of-the-art methods. Our code and dataset are publicly available at https://github.com/suchenjm/AutoThread.

Deep Reinforcement Learning (RL) is notoriously sample inefficient. One contributing factor is that RL agents are typically initialized from scratch, forcing them to acquire task-relevant knowledge through online interaction. Existing approaches obtain informative initializations through pre-collected datasets, high-fidelity simulators, or meta-learning over related tasks, but these prerequisites may be difficult to access or even unavailable. In this paper, we propose Programmatic Dynamics Priors for Value Network Initialization (ProDVI), a framework that leverages the commonsense and domain knowledge encoded in large language models to initialize RL agents without relying on these resources. Specifically, ProDVI prompts a code-generating language model to produce executable Python functions that encode coarse hypotheses about environment dynamics. These functions are then used to generate synthetic transitions. Based on these transitions, we construct an auxiliary dynamics prediction objective to pretrain the state-action encoder of the value network in an actor-critic framework. The learned representation provides dynamics-aware inductive biases before online RL begins. Importantly, the generated programs are used only for representation pretraining and are not required to faithfully simulate the target environment. While the generated programs may be inaccurate, their induced initialization can be corrected through online learning from real transitions and rewards. Experiments on OpenAI Gym and DeepMind Control Suite tasks show that ProDVI can effectively improve the sample efficiency of model-free RL algorithms.

Retail investors lack access to the kind of personalized, tax-aware portfolio management that institutional clients take for granted -- existing robo-advisors use static, rule-based allocation, and institutional-grade systems require account minimums and technology stacks unavailable to individual investors. We present a fully built, integration-tested application that closes this gap: a FastAPI backend and web dashboard that let a user describe an investment goal in plain language (e.g. "I want steady growth but need to sell some shares next month for a down payment"), routes that goal to one of six investment mandates, and produces a live, broker-integrated portfolio recommendation from athree-phase reinforcement learning system -- a self-supervised cross-asset encoder, a Mixture-of-Experts (MoE) allocation policy with a learned intent router, and a lightweight LoRA adapter that personalizes recommendations from an individual's revealed brokerage behavior without retraining the shared model. The system is functionally complete and integration-tested end-to-end against a live brokerage API (Alpaca, paper-trading mode), including multi-user authentication, a trust first preview-before-apply confirmation flow, daily email digests, and an auditable action-integrity chain, but has not yet been opened to real end-users; we report this honestly as an emerging, pre-deployment application with a concrete path to full deployment, alongside 14-day walk-forward backtests (bootstrapped confidence intervals included) as preliminary, pre-deployment validation rather than production performance. We also report several practical engineering lessons -- silently-inactive integration paths, hanging third-party API calls, and the value of end-to-end empirical verification over trusting checkpoint metadata -- that we believe generalize to other applied RL systems built on external, live data sources.

Autonomous cyber defense systems based on Deep Reinforcement Learning (DRL) have attracted significant research attention, yet remain evaluated almost exclusively against static, heuristic red agents, leaving their robustness against adaptive threats critically understudied. Meanwhile, recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have improved LLM reasoning, but their integration into cybersecurity remains elusive due to the absence of suitable benchmark environments and interaction datasets. To bridge this gap, we introduce Trident, an agentic LLM red teaming framework comprising three components: a dynamic benchmark with isolated sandbox servers spanning CybORG CAGE 4 and CyberWheel, a dataset comprises over 13,000 high-fidelity red-blue interaction trajectories for RLVR, and a ``Code-as-Policy'' RLVR agentic architecture Trident Agentic). The latter reformulates red agent training as a contextual bandit via a tripartite Log Summarizer--Planner--Coder design, where a trainable Planner generates complete attack strategies from compressed execution logs, which a frozen Coder translates into executable Python policies deployed against live DRL defenders. Empirical evaluations reveal a fundamental brittleness in existing defenses: with a single trainable 7B planner, Trident reduces blue agent defensive performance by an average of 522% compared to static red agent baselines while autonomously discovering emergent behaviors such as decoy avoidance and adaptive state prioritization that static heuristics entirely fail to uncover.

Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) exhibit fundamentally different behaviors in enhancing multi-task reasoning for large language models (LLMs). Our preliminary experiments revealed a phenomenon: SFT suffers from severe task conflicts under multi-stage training, whereas RL enables stable coexistence across diverse tasks. Empirically, we trace this to the parameter level, observing that RL induces sparse and approximately orthogonal updates across tasks. We provide a theoretical explanation for this mechanism by analyzing multi-task gradient interference. Our results reveal a distinction: interference in SFT is norm-limited, scaling with the absolute gradient magnitude, whereas interference in RL is variance-limited, bounded by the gradient variance induced by advantage normalization and on-policy optimization. This small variance bound yields near-orthogonal optimization directions across tasks. Leveraging this insight, we propose Parallel-RL, a paradigm that decouples multi-task training, significantly improving efficiency and flexibility.

Test-time reinforcement learning (TTRL) improves the reasoning capabilities of large language models without labeled data by updating the policy with pseudo-labels constructed through majority voting. While effective, the reward signal assigned from majority voting is highly sensitive to consensus strength, defined as the frequency of the most common answer within a rollout group. In TTRL, consensus strength plays a dual role: it reflects both the reliability of the pseudo-label and the distribution of advantages. Low consensus can amplify updates from unreliable pseudo-labels through disproportionately large advantages, whereas high consensus reduces reward contrast and ultimately yields vanishing gradients. In this paper, we introduce Hi-TTRL, a test-time reinforcement learning framework that utilizes hints during sampling to regulate rollout consensus strength. Hi-TTRL first estimates consensus strength from a partial rollout group. When the consensus strength falls outside a target interval, it invokes a Markov chain Monte Carlo (MCMC) hint sampler. The sampler targets the power-transformed prefix distribution and uses finite-step approximate sampling to generate rollout prefixes as hints. By tuning the power exponent, Hi-TTRL generates hints with a sharpened or flattened power target, steering rollout consensus strength toward the target interval. Experiments on multiple datasets and backbones show that Hi-TTRL consistently improves over standard TTRL, with ablations and consensus-steering analyses validating the effectiveness of adaptive hint-guided consensus regulation.

Deep reinforcement learning (DRL) approaches for flexible job shop scheduling (FJSP) heavily rely on attention-centric architectures to achieve state-of-the-art performance. However, these models suffer from excessive parameter counts and prohibitive inference latency as problem scales expand. While liquid neural networks (LNNs) offer a parameter-efficient alternative for modeling adaptive state evolution, their inherently sequential dynamics bottleneck computational efficiency. To resolve this trade-off, we propose PLAN (Parallel Liquid-inspired Approximation Network), a lightweight representation learning framework that reformulates continuous liquid-state dynamics into a discretized and parallelizable formulation. PLAN structurally decouples state evolution from context aggregation, where liquid-inspired updates handle the primary evolving state representation, and a lightweight context aggregation module provides complementary global context. Furthermore, PLAN acts as a versatile, plug-and-play backbone that generalizes to complex FJSP variants, pairing with a compact stochastic module for stochastic FJSP and replacing heavy heterogeneous graph transformers in multi-faceted dynamic FJSP. Extensive evaluations across deterministic, stochastic, and multi-faceted dynamic FJSP benchmarks show that PLAN reduces the average makespan by 1.2%, 1.4%, and 2.3%, respectively, compared with the corresponding state-of-the-art baselines, with the improvement reaching 10.2% in one benchmark setting. PLAN also reduces average inference latency by 13.2%, 31.7%, and 26.9%, respectively, with a maximum reduction of 69.2% on the largest instances, while using only 22$-$47% of the baseline parameters.

Curling is often referred to as "Chess on Ice", owing to the tactical complexity of its decision-making process. Yet unlike chess, curling remains largely underexplored from a machine learning perspective, with prior work confined mainly to statistical approaches. We propose a reinforcement learning framework capable of quantitatively evaluating and comparing tactical options in curling. The game poses several modeling challenges: continuous state and action spaces, stochastic action outcomes reflecting player skill variability, and state transitions that are highly sensitive to small perturbations in the executed action. To address them, we employ the Deep Deterministic Policy Gradient actor-critic algorithm, adapted to exploit the finite-horizon structure of the game. Our experiments show that effective curling strategies can be acquired in a fully self-supervised manner, without any human-annotated data: on a reduced four-rock variant, the learned agent matches a hand-crafted expert heuristic in a regime where that heuristic is close to optimal, a parity we quantify against the intrinsic hammer advantage of the variant. Beyond the resulting policy, the learned critic provides a dense value estimate over the entire continuous action space, enabling the quantitative comparison of tactical alternatives for applications such as post-game performance analysis and decision support during athlete preparation.

Many operational problems are constrained sequential decision processes with large, combinatorial action spaces and interdependent feasibility constraints. Mixed-integer linear programs (MILPs) handle such constraints flexibly but scale poorly in stochastic environments. Deep reinforcement learning (DRL) promises scalable decision rules, but existing methods either penalize constraints rather than enforce them, or rely on feasibility mechanisms that break down once constraints interact. We bridge this gap by embedding a differentiable convex optimization module inside the policy: a neural network proposes continuous action targets, a quadratic program projects them onto the relaxed feasible set, and a dual-informed integer mapping restores integrality while preserving feasibility. Given a differentiable simulator, the policy trains end to end from sampled trajectories using pathwise gradients, while handling hard constraints with similar flexibility to MILPs. We show that our feasibility enforcement has bounded error relative to an exact integer projection and ensures the entire feasible action space is reachable. We apply the method to multi-echelon production-inventory planning under shared resource and material constraints. Our policy attains an average optimality gap below 1% on small instances. It further outperforms state-of-the-art echelon base-stock policies by up to 9.75% and a rolling-horizon multi-stage stochastic program by at least 7.7% in larger networks. On an industry-scale case study from ASML, it reduces average cost by up to 3.22% relative to the best-known benchmark policy. The savings are largest where planning is hardest: in tightly capacitated systems with high demand variability. More broadly, our work shows that DRL can deliver economically significant savings in sequential decision problems with interdependent hard constraints, which are widespread in practice.

Executing long-horizon tool invocations in real-world environments is severely challenged by dynamic user intent noise. Existing methods attempt robustness via implicit history scanning or text compression, yet predominantly assume perfect instructions in simplistic scenarios. Inevitably, under fluctuating contexts, obsolete constraints dilute model attention, triggering catastrophic intent deviation and infinite API loops. To resolve this, we propose IACM-RL, a comprehensive framework for robust tool invocation. First, we introduce the DynamicIntent pipeline, synthesizing trajectories across 13 fine-grained fluctuation scenarios, paired with a five-dimensional diagnostic metric suite. Second, IACM-RL deploys a BeliefState-based Self-Generated Context Manager that proactively tracks shifting goals and isolates overwritten parameters using structural stale flags. To autonomously internalize this state-tracking capability, we optimize the policy using a hierarchical intent-driven reward alongside three auxiliary losses (action calibration, CM extraction, and state distillation). Experiments on DynamicIntent, BFCL-V3, and $\mathrmτ^2$-Bench demonstrate that IACM-RL significantly outperforms baselines, reducing infinite loops and stale context errors while enhancing out-of-domain generalization.

Online education offers unprecedented scalability and accessibility to global learners from diverse backgrounds, but it often suffers from low engagement and poor long term learning effectiveness. To address these challenges, we introduce AI Tutor, a reinforcement learning based model designed to promote sustainable learning by optimizing both short and longterm learning outcomes. In the short term, AI-Tutor draws on cognitive theory to guide learners through a balance of acquiring new knowledge and reinforcing prior learning. In the long term, it models learner engagement to inform strategies that sustain motivation and reduce dropout. These enhancements enable AI-Tutor to provide personalized guidance that fosters both effective learning and sustained participation. Empirical evaluations on 23 million learning records from 33,700 learners show that AI Tutor consistently outperforms state-of-the-art baselines across engagement, knowledge retention, and final learning outcomes. Learning path analyses further reveal how AI-Tutor adapts its strategies to learners with diverse profiles, offering adaptive and human-centered support.

Outcome-based reinforcement learning enables search-augmented language agents to learn from verifiable final answers, but its trajectory-level credit cannot distinguish the contributions of individual actions in a multi-turn search process. We propose EviSD, an evidence-conditioned self-distillation framework that uses instance-level supporting evidence as privileged information for search actions and golden answers as complementary privilege for answer actions. During training, the student samples actions from the original context, while the same model re-scores them as a privileged teacher under an action-aligned context. EviSD converts the detached teacher--student gap into a bounded correction to the outcome-derived GRPO advantage and applies it only to generated action spans. This design localizes privileged guidance while preserving the update direction determined by the outcome reward, without an auxiliary distillation objective or any change at inference time. Across seven question-answering benchmarks and three backbones spanning model scales and generations, EviSD achieves the highest macro-average Exact Match in all evaluated settings, outperforming the strongest compared methods by 1.3--2.3 points while modulating only 6.7%--15.1% of response tokens. Code is available at https://github.com/JiananXie/EviSD.

Adapting a pretrained vision-language-action (VLA) policy to a new robot usually assumes embodiment-specific demonstrations. This assumption is especially restrictive for custom robots whose morphology differs strongly from the manipulators seen in large robot datasets. We study a harder setting: zero-demo embodiment alignment of OpenVLA-OFT on a cable-driven parallel robot (CDPR) with a simple gripper and a previously unseen control interface. Instead of supervised fine-tuning, we use reinforcement learning in simulation with dense geometric rewards computed from simulator state. The training is performed in two stages: a PPO stage for directional motion primitives, followed by GRPO continuation from the PPO checkpoint with an expanded instruction space that includes object-conditioned commands. On the four shared directional instructions, the average held-out success rate improves from 34.25% after PPO to 53.50% after PPO$\rightarrow$GRPO, with especially large gains on \texttt{move left} and \texttt{move backward}. In the GRPO stage we additionally introduce \texttt{move to <object>} over eight target objects and obtain 39/400 = 9.75% strict success, while qualitative rollouts frequently show correct target-directed approach behavior before late-stage instability. Compared with prior OpenVLA and OpenVLA-OFT results, which rely on demonstration datasets and mostly standard rigid-arm embodiments, our method uses no embodiment-specific dataset at all. The results do not yet establish robust manipulation, but they provide stronger evidence that RL-only bootstrapping can create the first usable language-conditioned controller for a genuinely novel embodiment.

Existing search-augmented LLM agents are trained using Reinforcement Learning to boost its reasoning capabilities. However, these approaches primarily rely on outcome-level rewards, which provide little supervision over search behavior and overlook agent's ability to decompose complex queries properly. To mitigate this issue, we propose PROGRESS which utilizes teacher-guided coverage reward to explicitly shape decomposed query generation of the policy model. During training, frozen teacher models are used to decompose complex queries into essential search queries. These essential search queries are utilized to guide the search behavior of the policy model. Integrated into an R1-style training framework, our approach provides lightweight guidance over query decomposition decisions without dense process-level supervision. Experiments show that coverage-guided RL improves overall task performance, highlighting the importance of explicitly supervising search strategies in agentic LLMs.

Microrobots hold significant potential for various applications, where targeted navigation is a basic requirement. Deep reinforcement learning (DRL) has recently emerged as a powerful paradigm for fully autonomous microrobot navigation. Yet, current DRL-based approaches pay limited attention to learning efficiency and effectiveness, requiring hours to days for model training. Consequently, this impedes both rapid practical deployment and parameter optimization. To address these challenges, we present a learning framework that enables effective microrobot navigation policies to be trained within minutes. In the proposed framework, we develop a fully vectorized simulator with more than 10,000 artificial vascular environments, parallelizing dynamics, LiDAR-inspired perception, and feasibility checks across thousands of environments to achieve roughly 190,000 transitions per second. To achieve effectiveness in fast training, we propose a task-shaping-regularization (TSR) reward framework. The TSR framework accelerates convergence, improves final performance, reduces action variation by at least 33.7%, and increases obstacle clearance by at least 2.1% across all evaluated scenarios. Results show that the proposed learning framework reduces training time to under 10 minutes, while supporting zero-shot deployment across distinct microrobot types and navigation scenarios. Collectively, this framework can substantially shorten the design loop and accelerate the deployment of autonomous microrobots.

Compact web agents can reduce deployment cost, but training them poses challenges in both data collection and post-SFT reinforcement learning (RL). Successful trajectories are expensive to collect and often contain inefficient detours. After supervised fine-tuning (SFT), full trajectory corpora are dominated by routine states; moreover, when group-relative RL is applied to web actions, inadequately designed action-level rewards can yield weak or misleading relative updates, while groups rejected as unsuitable for such updates receive no fallback learning signal. We present RMSWeb, a three-part recipe for Qwen3-VL-Instruct at 8B and 32B. Reflection-conditioned retries increase collection yield and shorten successful trajectories; failure-mode mining concentrates offline RL on critical states exposed by the SFT policy; and Salvage-DS combines an action-semantic polarized reward, contrast-and-competence-gated dynamic sampling, and an action-only anchor for rejected groups. Policies trained with reflection-collected data use up to 19.7% fewer action steps on solved tasks. On WebVoyager, Online-Mind2Web, and WebTailBench, RMSWeb improves over SFT by 2.4-7.0 points at 8B and 1.2-7.7 points at 32B. Our 8B model also achieves the strongest reported Online-Mind2Web result among similarly sized open-weight models in our comparison and a leading reported accuracy-cost trade-off on WebVoyager and WebTailBench, with the caveat that external evaluation protocols differ.

Error-penalized scoring rules ($+1$ for a correct answer, $-λ$ for a wrong one, $0$ for abstaining) are increasingly prescribed against hallucination: a rational agent facing such a rule answers exactly when its correctness probability exceeds Chow's threshold $t^\ast=λ/(1+λ)$. We prove that a KL-anchored gradient learner can do the opposite. When abstention is a discrete action, the reward gradient and the anchor's restoring force are throttled by the same gate-saturation factor and die together: under explicit conditions (among them, blanket answering loses score in expectation and prompts share a bounded readout) the model drifts toward refusing everything, its mean training reward rising to zero like $1/t$ in training time $t$, so the curve reads as improvement while coverage collapses. The advantage estimator compounds the failure: in its sparse-answer regime, group normalization silently replaces every designed penalty with an effective penalty of one, moving the learned threshold from $t^\ast$ to $1/2$. The repair is structural: train a mandatory confidence report with a strictly proper score plus a correctness reward, and abstain only at deployment by thresholding the report. The always-emitted report has no gate to saturate, so no shared factor can kill its reward gradient and its anchor together, and its calibrated optimum is attracting. Simulations confirm every prediction, and experiments on language models at two scales confirm the mechanism live: the rule silences questions the models demonstrably still solve within ten optimizer steps, an ablation isolates the cause, and report-level training raises coverage, accuracy, and calibration together.

We show that on-policy reinforcement learning with verifiable rewards (RLVR) can improve the current objective while making successful behaviors for later objectives too rare to sample and reinforce. We call this verifier-induced support reshaping and define effective rewardable support as successful trajectories reachable within a fixed rollout budget. Across two model families, we study this effect through repeated verifier-scored sampling and bidirectional training on mathematical reasoning and constrained instruction following, including sequential training with the opposite verifier. Math-RLVR raises average instruction-following success but reduces the number of prompts with any successful response under repeated sampling. On IFEval with Qwen3-8B-Base, pass@1 rises by 6.5 percentage points while best@32 falls by 9.8 percentage points, and the same divergence appears across both models and IF benchmarks. Conversely, IF-RLVR shifts math responses from step-by-step openings toward direct answers, lowers best@k across sampling budgets, and reduces reward variation for later Math-RLVR. Token-distribution analyses and controlled opening interventions show that these changes concentrate in the first few response tokens. RLVR mainly reranks openings already available in the base policy, and the selected opening causally affects math searchability. The tested reference-policy constraints, routing priors, and on-policy distillation preserve cross-task support only partially; MathIF and ReasonIF show that marginal gains translate only partly into responses that are both correct and constraint-following. Therefore, endpoint improvements do not guarantee future trainability or joint capability under on-policy optimization. Code is available at https://github.com/sylvain-wei/verifier-induced-support-reshaping

Conflicting objectives are general in RL alignment, and training on them data-efficiently is hard. Training a safety guard with RL means optimizing two objectives that conflict: catch real harm, and do not refuse benign prompts. Our finding is that over-refusal improves 22.4% to 12.8%, while under-refusal on adversarial attacks silently worsens 0.27 to 0.33. We present C-Guard, a constitution-grid instrument that generates the RL training data, and C-LIM, a per-cell learnability score that decides each cell's move: prune, densify, amend, expand. C-LIM flags the dead-weight data region before any training budget is spent: 187 untargeted rows had bought zero gain, and our method lifts the same region's learning impact 0.733 to 0.80. Code and the constitution are open-sourced.

Deep reinforcement learning (RL) agents achieve strong performance by optimizing scalar reward functions. However, once deployed, the policies of these RL agents are often rigid and costly to adapt to new performance criteria. For instance, an agent trained to maximize expected cumulative reward may not accommodate previously unknown stakeholder preferences. Existing approaches to achieve fairness, a type of preference, in RL typically assume that such preferences are known a priori and require complete retraining of the policy under a fairness-oriented metric. Inspired by inference-time alignment in large language models, we investigate the problem of steering a pretrained RL policy toward welfare-based fairness objectives at inference time without updating the base policy's parameters. We formalize inference-time fairness alignment as a policy shaping problem and propose a multiplicative policy shaping framework that adjusts action probabilities using action-dependent welfare scores, thus requiring no modification to the base policy. Our framework is general and compatible with any deep RL agent. Through extensive experiments across multiple domains, we demonstrate that inference-time policy shaping substantially improves welfare-based fairness objectives while preserving core task performance.

Cable-driven parallel robots (CDPRs) present diverse configurations and complex control challenges, which can be addressed by deep reinforcement learning (DRL) by learning their nonlinear dynamics. However, DRL methods often require extensive training time, and the resulting policies do not generalize well to different robot configurations or varying numbers of actuators. In this article, we introduce a novel DRL approach for controlling CDPRs that does not depend on the specific robot configuration. Our method trains an actuator-level policy that controls each motor to achieve its target cable length, in contrast to conventional DRL approaches that learn to control the entire robot to reach a desired end-effector position. To the best of our knowledge, this is the first work to apply DRL to control CDPRs using an actuator-level policy. This approach offers two main advantages: (i) a single shared policy can be applied to any CDPR configuration, regardless of actuator count, and (ii) reliance on inverse kinematics, avoiding the more challenging forward kinematics problem. Training is performed in simulation, and the learned policy is successfully transferred to a real CDPR. Experimental results show that the actuator-level policy (ALP) surpasses traditional reinforcement learning methods in both robustness and precision. We further control a real 8-motor CDPR with 3D motion using a policy trained on a simulated 4-motor planar CDPR operating in 2D. This illustrates that the proposed method is applicable to any CDPR configuration, independent of actuator number or placement.

In Formula 1, drivers optimize racing lines within tire grip limits to minimize lap times; however, in rally racing, drivers intentionally break traction to drift on loose surfaces. This maneuver rapidly aligns the vehicle for corner exits, ultimately reducing lap time. Autonomously executing such maneuvers formulates a complex dual-objective control problem: stabilizing highly nonlinear drift dynamics while strictly minimizing lap time. Addressing this challenge motivates the development of advanced Minimum-Lap-Time (MLT) drift control architectures. This paper proposes a planning-control framework specifically designed for MLT drifting scenario. First, we formulate an optimal control problem to generate a MLT drift planning trajectory, which is used as prior data to train a deep reinforcement learning drift controller. Given that drifting involves extremely large sideslip angles and is therefore challenging to learn directly, a Track-guided Reinforcement Learning (TgRL) drift control method is proposed to enable progressive training in a step-by-step manner, from drift control policy, to drift corner policy, and finally to a comprehensive drift race policy. The reward function incorporates both an instant reward term and an end reward term derived from the Minimum-Lap-Time objective. Simulation results demonstrate that the proposed framework enables the agent to learn a drift racing policy that not only ensures vehicle motion control performance but also effectively reduces lap time.

The fundamental goal of agentic visual reasoning is to improve the success rate of multimodal large language models (MLLMs) on complex tasks, rather than merely equipping them with a sophisticated yet inefficient reasoning paradigm. In this work, we rethink agentic visual reasoning through two key dimensions of tool use: Mode Adaptiveness (MA) and Tool Effect (TE). Mode Adaptiveness characterizes whether an MLLM can recognize when tools are truly necessary and invoke them accordingly, thereby avoiding unnecessary computational overhead while improving performance on challenging problems that require tool assistance. Tool Effect characterizes the actual impact of tool use: tools should extend the model's capabilities on problems unsolvable through text-only reasoning, while avoiding additional errors on problems that the model can already solve without tools. We conduct a comprehensive analysis to quantify these two properties and empirically reveal that existing agentic visual reasoning models exhibit limited Mode Adaptiveness, while the gains produced by tool use on hard examples are largely offset by the harm introduced on easy examples that the models can already solve. Motivated by these observations, we propose Beacon, a novel agentic visual reasoning model that achieves stronger overall performance, improved Mode Adaptiveness, and genuine tool-induced performance gains. At the core of Beacon are the Necessity-Aware Adaptive Reward and the Hint-Guided Capability Expansion mechanism in the reinforcement learning stage, which respectively encourage adaptive tool invocation based on task necessity and strengthen the model's tool-use capability on the most challenging problems. Extensive experiments across diverse benchmarks demonstrate the strong overall performance of Beacon and its substantial improvements in both Mode Adaptiveness and Tool Effect.

Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifiable task environments with execution feedback (OpenMLE-Gym), operator learning (OpenMLE-RL), and long-horizon search (OpenMLE-Evo). On this stack we post-train Frontis-MA1 (35B) as a meta-evolution agent for MLE, aligning post-training and inference around four atomic program-evolution operators (Draft, Improve, Debug, Crossover): the same operators are trained via execution-grounded SFT and RL on data deduplicated against all evaluation benchmarks, then composed into long-horizon search, coupling learning and evolution in a single loop. On MLE-Bench Lite under a 12-hour per-task budget on one RTX 4090 capped at 12 GB VRAM, Frontis-MA1 (35B) improves Medal Average from 39.39% to 60.61% over its base model with OpenMLE-Evo, and reaches 71.21% with OpenMLE-Evo-Max (benchmark-independent experience priors and asynchronous search), exceeding GPT-5.5 + Codex and approaching GPT-5.6 Sol and the 2.8T Kimi K3. On held-out NatureBench Lite, both components transfer: with the framework fixed, swapping in the trained model raises Match-SOTA from 50% to 70%; with the model fixed, swapping in OpenMLE-Evo raises it from 20% to 50%. We release the model weights and the full OpenMLE stack to enable reproducible research on executable AI4AI toward RSI. Code: https://github.com/FrontisAI/OpenRSI

Detection and localization of AI-tampered images are critical for trustworthy AI, yet modern generative models have made such manipulations increasingly difficult to identify. While traditional binary classifiers can detect image tampering, they lack interpretability and generalization. Vision-Language Models (VLMs) offer a promising alternative due to their strong visual understanding and reasoning capabilities; however, existing approaches typically rely on supervised finetuning with curated explanations rather than exploiting their inherent reasoning capabilities. In this work, we investigate whether VLMs can be trained to reason about AI-generated image edits using reinforcement learning (RL) rather than explicit reasoning supervision. Motivated by the success in Group Relative Policy Optimization (GRPO), an RL technique that incentivizes the model to reason by generating thinking traces prior to giving the final answer, we propose a GRPO-based training framework that utilizes simple accuracy and format rewards. Given an input image, the model produces a structured reasoning trace and predicts whether the image has been tampered with. A lightweight segmentation model is then guided by the reasoning output to generate pixel-level localization masks. Experiments across multiple image manipulation datasets demonstrate that our approach achieves competitive detection and localization performance compared to state-of-the-art image forgery detectors, despite requiring substantially weaker supervision. We introduce effective intersection over union (eff-IoU), a unified metric to jointly evaluate detection and localization. These results suggest that reinforcement learning provides an effective and scalable mechanism for teaching VLMs to reason about AI-generated content.

Scaling test-time computation can improve language-model reasoning, but uniform budgets waste computation on easy inputs, while verifier-guided refinement relies on external feedback. We introduce Self-Verifying Refinement (SVR), an oracle-free multi-turn reinforcement learning framework that learns to use self-verification as a compute-control policy. At each turn, the model produces a solution together with a discrete correctness verdict and a confidence score; it retains the current answer only when the verdict is Correct and confidence exceeds a threshold, and otherwise continues refinement using its own self-verification. Ground-truth correctness is used only to construct training rewards and is never exposed to the policy through refinement prompts or required at inference. SVR is trained with GRPO on fixed-horizon trajectories using rewards that promote solution correctness, calibration-aware self-verification, and stop-ready correct states; adaptive stopping is activated only at inference. On seven mathematical reasoning benchmarks with Qwen3.5-2B, SVR achieves a macro-average accuracy of 0.563 with only 2.99 inference turns on average. In the evaluated complete-system comparison, it exceeds standard GRPO, strong multi-turn baselines, and a fixed-budget oracle-guided score-feedback reference while requiring substantially fewer turns than fixed ten-turn inference. These results demonstrate that learned self-verification can serve as an effective internal control signal for answer retention and adaptive test-time compute allocation.

Biomedical laboratory robots must navigate to instruments before performing experimental procedures. Existing embodied navigation platforms are designed for household environments and treat a target as an object center or an arbitrary nearby position. This representation is inadequate for laboratory instruments, which must be approached from their operating side while maintaining safe clearance from surrounding equipment. We introduce BioVLN, a simulation platform for developing and evaluating visual-language navigation agents in biomedical laboratories. BioVLN represents each instrument with three regions: its physical body, a surrounding clearance region, and an operation area in front of the usable side. This model is applied consistently to scene generation, target placement, navigation evaluation, and safety analysis, so success depends on reaching a position from which the instrument can be accessed. BioVLN supports procedural scene generation and manually designed environments, producing 47 scenes and 1667 episodes. Standardized navigation and reinforcement-learning interfaces enable trajectory collection and policy training. Experiments show that geometric exploration reaches 74.4--87.5% success, while sampling multiple valid positions in the operation area improves success to 83.3--92.5% and reduces unsafe proximity.

Test-time reinforcement learning (TTRL) enables language models to self-evolve at inference time without labeled feedback. Existing methods rely on answer voting and therefore do not extend naturally to open-ended generation, where valid responses cannot be mapped to a shared canonical answer. Without external reward models or stronger judges, adaptation must instead construct reliable rewards from the model's own outputs. We introduce SERPO (Self-Evolving Rubric Policy Optimization), which replaces answer voting with a closed loop that co-evolves response evidence, query-specific rubrics, and policy parameters. Good-Normal-Bad (G-N-B) response evolution organizes maximally separated rollouts into ordered archives; rubric evolution retains criteria that discriminate these archives; probabilistic criterion scoring converts verdict-token likelihoods into reward signals; and policy evolution optimizes the actor with the resulting signals. New actor rollouts then refresh both the archives and rubrics, closing the three-way evolution loop. Across two model configurations, two in-domain benchmarks, and four OOD benchmarks, SERPO improves HealthBench and ResearchQA by up to 20.63 and 20.31 points over the corresponding base models, raises the six-benchmark macro-average by up to 8.06 points, and supports OOD transfer and continued cross-benchmark evolution.

Unmanned aerial vehicles (UAVs), particularly quadcopters, present unique challenges for autonomous control due to their underactuated dynamics: only four available control inputs must govern six degrees of freedom. This paper investigates a physics-aware, end-to-end deep reinforcement learning (DRL) approach that acts directly on low-level body inputs, total thrust and body torques $(T, τ_x, τ_y, τ_z)$, and closes the loop through a high-fidelity Simulink environment. Our simulator integrates a 12-state rigid-body model (MATLAB Level-2 S-Function) with (i) an Action2RPM allocation based on the Moore-Penrose pseudo-inverse of a coefficient matrix derived from thrust and drag terms, and (ii) first-order actuator dynamics for each motor (time constant $T_m = 0.076$ s), including rotor gyroscopic coupling. A shaped reward balances goal-reaching and stability using an exponential position well, attitude penalties, and quadratic velocity costs. Four DRL algorithms, DDPG, TD3, PPO, and SAC, are evaluated in two stages: (S1) thrust-only hover and (S2) hover with pitch torque and a translated goal. Results show that SAC and TD3 achieve superior stability and exploration efficiency, while PPO is less sample-efficient. The study highlights the significance of modeling actuator lags and aerodynamic moments for stable low-level control and provides a reproducible benchmark for quadcopter DRL.

Deep reinforcement learning (DRL) agents achieve strong performance in complex environments, yet their decision-making processes remain difficult to interpret. We introduce SPOT (Sampling Policy Observation Tree), a novel model-agnostic, sampling-based framework for interpreting DRL policies. Given access to the policy and an environment simulator, SPOT constructs an interpretable finite-horizon tree by sampling actions and recursively simulating the resulting successor states. The tree provides an empirical representation of the policy's action preferences and their possible downstream evolution. We provide formal guarantees establishing SPOT's asymptotic recovery of the policy's unique most probable action and characterizing its disagreement behavior under high-entropy policies. We demonstrate SPOT in the SUMO-RL traffic-signal control domain. The case study illustrates how its tree-based representation can be used to inspect policy preferences, compare alternative future trajectories, and reveal downstream behaviors that are not visible through single-timestep feature-attribution methods.

Experience replay remains one of the most practical and useful algorithmic tools in the deep reinforcement learning (DRL) toolbox. Aside from the limited success of prioritized replay and specialized approaches for large asynchronous systems, most DRL algorithms make use of a large, uniformly sampled recency buffer---even the size, one million, remains unchanged. Could we store less data, reduce redundancy, or more effectively chain experience together to speed up value propagation and still retain the performance of large buffers? In this paper, we investigate a simple compression approach that stores representative transitions derived from the end-points of a chain of connected $n$-step sequences. By curating these end-points in a smaller recency buffer, our method maintains an effective memory horizon comparable to a standard large buffer while requiring an order of magnitude less storage. Through empirical evaluation, we demonstrate that this approach prevents the systematic bias inherent in naive compression strategies and matches the performance of traditional large buffers in the Pinball environment and the Atari 2600 benchmark.

In safety-critical sectors such as robotics and automotive engineering, the deployment of Deep Reinforcement Learning (DRL) is often hindered by the black-box nature of deep neural networks. This lack of transparency poses significant challenges for regulatory compliance and human-agent trust. This paper presents an experimental study aimed at making high-performance continuous control DRL systems interpretable. A policy distillation framework is implemented using the classic Inverted Pendulum benchmark. A high-performance Twin Delayed DDPG (TD3) agent serves as an opaque, continuous teacher model, whose policy is distilled into an interpretable student surrogate based on a shallow Decision Tree. By leveraging a custom physics-aware feature and "Noisy Oracle Rollouts" for dataset generation, the distillation process achieves performance equivalent to the expert teacher. Furthermore, comparative control theory analysis reveals a fundamental trade-off: transitioning from continuous to discrete rule-based control induces high-frequency Bang-Bang actuation and a stable bimodal limit cycle. Simulation results indicate that Bounded-Input Bounded-Output (BIBO) stability is maintained while providing both global and local interpretability for safe autonomous systems.

Agentic search enables large language models to solve knowledge-intensive tasks by interleaving multi-step reasoning with retrieval, yet optimizing this with outcome-based reinforcement learning (RL) provides only sparse supervision. Knowledge distillation can supply denser guidance, and advanced proprietary models with their strong reasoning capabilities are promising teachers. While distilling from proprietary models can densify this supervisory signal, conventional logit-matching is precluded by hidden logits and mismatched tokenizers, whereas raw natural language trajectory imitation transfers superficial stylistic artifacts rather than core reasoning competence. To address the heterogeneous distillation problem and bridge the distribution gap, we propose Multi-Agent Protocol Distillation (MAPD), a joint distillation and RL framework uses a structured, style-normalized protocol as an intermediate representation. An offline multi-agent system (MAS) decomposes each query, retrieves supporting evidence, repairs failed searches, and converts the resulting exploration trace into a JSON protocol containing the task type, reasoning plan, and extractive grounding facts. During training, the protocol is provided only to a privileged branch of the student policy, whose token distributions furnish a dense distillation signal alongside the sparse RL objective. Extensive evaluations across seven QA benchmarks demonstrate that MAPD consistently outperforms competitive distillation and RL, achieving average success rates of 39.4% on Qwen3-1.7B and 44.4% on Qwen3-4B. Crucially, the framework generalizes robustly across diverse proprietary teachers while effectively mitigating the student policy from style drift and verbosity degeneration.

Reinforcement Learning (RL) training for Large Language Models (LLMs) often suffers from instability due to the discrepancy between training and inference. This training-inference discrepancy stems from two primary factors: an architectural separation between training and inference engines, and the use of low-precision quantization in inference versus higher-precision computation in training. To address training instability issues caused by high training-inference discrepancy, we present the principles and methods for its adaptive control. We propose Adaptive Control Reinforcement Learning (ACRL), which adaptively maintains the training-inference discrepancy within a reasonable range to ensure stable RL training. Beyond stabilization, ACRL inherently increases policy entropy, thereby enhancing exploration and improving accuracy. The experimental results show that when the inference engine utilizes FP8 quantization, ACRL consistently maintains the training-inference discrepancy within a reasonable range and stabilizes RL training. Furthermore, ACRL not only matches the accuracy of the BF16 baseline but also outperforms importance sampling (IS) fixes.

Operating constrained dynamical systems requires controllers to efficiently solve complex tasks while enforcing recursive feasibility and safety constraints. To address these competing requirements, we present Feasible Action for Optimal Control (FAOC), a novel control framework integrating Reinforcement Learning (RL) and Optimal Control (OC). The key contribution is a computationally efficient, optimization-based mapping algorithm that transforms the RL agent's action from a static abstract set into a state-dependent feasible parameter set of the Optimal Control Problem (OCP), guaranteeing strict satisfaction of the dynamical system's constraints. Thus, FAOC effectively combines the predictable safety of OC with the flexibility of RL. In contrast to prior work, the abstract action space of the RL agent does not require expert or heuristic design, and the OCP formulation is not compromised by the inability of RL to guarantee feasibility. We apply our approach to real-time motion planning for robot table tennis, which encapsulates these challenges. Via simulated experiments, we show that FAOC outperforms state-of-the-art baselines in both sample efficiency and closed-loop performance.