Deep Reinforcement Learning

Recent work trains reasoning models with length penalties to curb overthinking and cut inference cost. We show that these penalties make the chain of thought less monitorable. A length-compressed model still lets misleading hints steer its answers, but it less often verbalizes their influence. We train Qwen3-4B and Qwen3-14B with reinforcement learning under length penalties targeting 60% down to 30% of baseline chain-of-thought length, then evaluate them with nine types of biasing hints on held-out MMLU-Pro-R and four transfer benchmarks. A chain is faithful when an LLM monitor can tell from it that the hint influenced the answer. At the 30% target, accuracy stays near baseline and wrong-answer hints switch answers as often as before. Yet faithfulness drops on every evaluation set for both models, by 39% for Qwen3-14B and 35% for Qwen3-4B on MMLU-Pro-R. A control trained with the same correctness and format rewards but no length penalty leaves faithfulness intact or raises it. Shortening alone does not explain the drop. Compressed chains mention the hint 7 to 35 percentage points less often than the uncompressed model's chains shortened to the same length by random sentence deletion, across both model sizes and all five evaluation sets. Length penalties therefore trade monitorability for inference cost by removing the evidence monitors depend on.

Reinforcement learning (RL) enables the synthesis of control policies directly from data, making it highly appealing for complex cyber-physical systems (CPSs) and robotics. A persistent challenge, however, is ensuring strict, hard safety constraints during the active learning phase. In real-world physical systems, violating mechanical limits can cause irreversible damage, necessitating that exploration remains strictly within safe operational regions. We propose a generalized framework that combines the adaptive, high-performance nature of deep reinforcement learning (DRL) with the formal safety guarantees of model predictive control (MPC). Using a mathematical model of the system dynamics, offline MPC computations define a feasible state-action space, representing all safe combinations of system states and control inputs that guarantee constraint satisfaction. During training and deployment, the RL agent's instantaneous actions are projected onto this globally verified feasible set via a safety filter. We systematically evaluate our generalized approach on a non-linear 1-DoF laboratory testbed, demonstrating successful exploration and stable policy convergence on physical hardware.

We present FootsiesGym, an open-source environment for learning in a non-trivial two-player, zero-sum, imperfect-information game. Built on HiFight's minimalist 2D fighting game Footsies, it isolates the cyclic, non-transitive strategic interactions of fighting game neutral play while remaining simple enough for efficient analysis. We provide a vectorized simulator that enables high-throughput training on standard hardware, making the environment accessible and reproducible. We describe the design of the environment, benchmark several reinforcement learning algorithms, and discuss open research directions it enables. The code is available at https://github.com/como-research/FootsiesGym.

Large language model (LLM) agents require post-training methods that can improve long-horizon decision making from environment feedback. However, existing agentic post-training pipelines often treat data curation as a fixed preprocessing step, focusing mainly on data augmentation while neglecting filtering, refinement, and adaptation to downstream failures. We propose CurateEvo, a failure-driven dynamic evolution framework for agentic post-training data curation. CurateEvo represents the curation strategy as executable code and iteratively rewrites it using failed trajectories from a held-out development set. At each epoch, the evolved strategy transforms a fixed raw corpus into supervised fine-tuning data, reinforcement learning data, and an inference-time memory bank. The evolution process first improves effectiveness by diagnosing recurring failure modes and augmenting, filtering, or refining data accordingly, and then improves efficiency by pruning redundant or low-utility training turns under a cost-aware objective. Experiments on ACEBench-Agent, BFCL-V4, and τ^2-Bench under both labeled and wild-data settings show that CurateEvo consistently outperforms prior curation methods, improving average scores by 3.2 and 2.7 points, respectively. Further analyses demonstrate that CurateEvo is compatible with different post-training recipes and substantially reduces curation overhead.

In this paper, we investigate whether a model-free RL agent can identify and exploit price manipulation opportunities more effectively than a traditional model-based approach that assumes correct specification of the data-generating process but relies on noisy parameter estimates. We consider a single-asset market in which prices evolve according to an Almgren-Chriss framework with non-linear permanent impact and linear temporary impact. We first establish the existence of price-manipulative strategies in discrete time and compute the optimal benchmark strategy using Sequential Least Squares Quadratic Programming under full information. We then compare two finite-sample learning approaches: a model-based procedure that estimates impact parameters from simulated execution data and an agnostic RL approach based on Deep Deterministic Policy Gradient, trained directly on the same amount of data. For intermediate volatility, the RL agent successfully discovers profitable manipulative strategies without explicit knowledge of the underlying model, even when training data are quite limited. More importantly, RL consistently outperforms the model-based approach when parameter estimates are affected by sampling error, despite the latter benefiting from the correct model specification. For large volatility, all methods are unable to identify manipulation opportunities, while for small volatility, the model based approach outperforms RL. These findings highlight both the effectiveness of RL in complex control problems and the risks associated with deploying learning algorithms in financial markets without appropriate safeguards.

Portfolio optimization under uncertainty is inherently a multi-objective decision problem involving complex interactions among return, risk, market dynamics, and practical investment constraints. Existing reliability based portfolio optimization approaches primarily rely on static optimization frameworks and often fail to capture sequential decision making, tail risk, and market frictions such as transaction costs. To address these limitations, we propose a deep reinforcement learning framework for multi-objective reliability based portfolio optimization (MORP-DRL). The proposed framework jointly optimizes expected return and downside risk using three complementary risk measures: variance, Conditional Value-at-Risk (CVaR), and Entropic Value-at-Risk (EVaR). To model uncertainty and heavy-tailed market behavior, asset returns are represented using GARCH(1,1), Extreme Value Theory, and a t-copula dependence structure, while realistic scenarios are generated through quasi-Monte Carlo simulation. A Proximal Policy Optimization (PPO) based strategy is developed under practical constraints including transaction costs and portfolio bounds, and is benchmarked against NSGA-II. Experiments on ten global equity indices across pre-COVID, COVID, and post-COVID market regimes demonstrate that MORP-DRL achieves competitive risk-return performance, reduced downside risk during periods of market stress, and scalability to high-dimensional portfolio settings.

Battery charging of Autonomous Mobile Robots (AMRs) in warehouses is a critical operational challenge that heavily impacts both order processing times and throughput. In this study, we address the dynamic AMR charging problem under stochastic order arrivals, where robots must learn optimal charging decisions. Traditional fixed-rule heuristics often prove suboptimal in dynamic environments and fail to account for multi-AMR coordination, leading to severe resource inefficiencies. To overcome these limitations, we propose a Proximal Policy Optimization (PPO)-based Deep Reinforcement Learning (DRL) framework designed for multi-block warehouses with fixed charging stations. Our model dynamically learns two key decisions: charging station selection and optimal charging duration, explicitly accounting for anticipated queuing times at the stations. Extensive numerical experiments benchmark the proposed model against state-of-the-art DRL and traditional heuristic approaches. Results demonstrate that our PPO framework increases order-completion rates by up to 6% compared to the strongest baseline, while significantly reducing the total time dedicated to recharging operations. Furthermore, we validate the model's robustness across diverse warehouse configurations and stochastic arrival rates. Finally, we interpret the learned DRL policy, offering valuable operational insights into its superiority over standard benchmarks.

Reinforcement learning with verifiable rewards (RLVR) is a powerful recipe for improving language-model reasoning, but it is expensive to repeat on every new strong model because the target model must generate many rollouts during training. As models scale, post-training itself becomes a bottleneck. We study a weak-to-strong alternative: run RL on a smaller model where rollouts are cheaper, then reuse what that RL run learned to improve a stronger target model. Directly distilling the post-RL weak teacher is not enough, because the teacher's final policy mixes useful RL gains with the limitations of the smaller model. We propose Direct On-Policy Distillation (Direct-OPD), which transfers the teacher's RL-induced policy shift instead. Direct-OPD compares the post-RL teacher with its own pre-RL reference and treats their log-ratio as a dense implicit reward for the student. In plain terms, the checkpoint pair tells us which actions RL made the weak model more or less likely to take, and Direct-OPD applies that signal on the stronger student's own on-policy states. This directly reuses the weak model's RL supervision signal without running sparse-reward RL on the target model. Empirically, Direct-OPD consistently leverages weaker teachers to improve stronger target models; notably, it boosts Qwen3-1.7B from 48.3% to 58.3% on AIME 2024 in just 4 hours on 8 A100 GPUs. It outperforms step-matched direct RL and enables the sequential composition of multiple policy shifts. Our results show that RL outcomes can be reused across model scales as implicit reward signals, not merely as final models to imitate.

In liberalised railway systems, operators must set prices dynamically in an environment with partial observability, as they retain private information about their objectives and performance, where regulatory constraints prohibit communication or direct information exchange between competitors to prevent explicit collusion. Consequently, agents must learn to infer strategic interactions only from observable market data which presents a significant challenge for multi-agent reinforcement learning, where standard approaches typically treat observations as unstructured vectors, ignoring the underlying market topology that governs strategic interactions. To address this, an entity graph modelling approach is proposed, which represents the environment as a graph of operational units, rather than decision-making agents or static infrastructure, encoding competition, coordination, and connectivity relations between entities. Then, an extension of the multi-agent twin delayed deep deterministic policy gradient algorithm with graph-based representation learning processes the features of the entities through a multi-layer relational graph convolutional network and aggregates them via a learnt attention mechanism. Experimental results in a rail pricing reinforcement learning environment show that this novel framework achieves higher revenue and stability in two different settings of increasing market complexity compared to a representative selection of relational and non-relational baselines. The code is publicly available at: https://github.com/Kinrre/RelationalRailPricing-RL

Vision-based deep reinforcement learning involves dealing with high-dimensional inputs of image information. It is crucial to abstract effective states from high-dimensional image inputs and limited samples for sample-efficient reinforcement learning. To address this challenge, inspired by fields such as natural language processing and computer vision, we propose a self-supervised task based on mask prediction as an auxiliary task for reinforcement learning. This non-reconstruction method uses the sequence information collected by the agent from the environment and the context information in the sequence to predict the masked information, thereby strengthening the agent's understanding of the task and learning effective representations. Combined with transformers, we find that the model reconstructs the masked input sequence in the latent space. By feeding the compressed representations learned by this method into reinforcement learning models, we observe an improvement in the sample efficiency of reinforcement learning. Moreover, the model outperforms state-of-the-art sample-efficient reinforcement learning methods on multiple continuous and discrete control benchmarks.

Current agentic workflows usually involve decomposing user requests into sequences of tool calls with correctly resolved parameters, the results of which are processed through reasoning traces in the language model's context window. The prevailing route to improve such reasoning is test-time scaling, which trains models to search over long chains of thought; but the resulting capability is entangled in model weights, is not verifiable step-by-step, and is costly at inference. We present Forethought, a neurosymbolic reasoning system that instead treats reasoning as an explicit, verifiable program, that builds from a library of symbolic and neural primitives which are composed through a domain-specific language. The result are reasoning programs, which are concrete representations of the model's work, and as such can be inspected and modified before deployment. Instantiated as a tool-calling execution kernel and evaluated across five benchmarks, Forethought improves base-model accuracy by about 30% relative and outperforms vanilla prompting, reinforcement learning scaffolds, and prompt-evolution methods, enabling small models to match or exceed frontier models capabilities. In a direct comparison, a non-reasoning model augmented with Forethought competes with a dedicated reasoning model while requiring roughly three orders of magnitude less post-training investment, and remains model-agnostic and auditable.

This study independently replicates and extends the Natural Language Tools (NLT) framework of Johnson et al.~(2025), which questions the use of structured tool calling in large language model (LLM) agentic systems. We evaluated NLT across 14 models and 8,560 trials, adding newer frontier, reasoning, and open-weight models to the original set. The results confirm the core findings and add detail. NLT improves tool-calling accuracy by 14.9 percentage points overall (62.3% versus 47.4% structured) and reduces critical errors by 93% (51 versus 755 errors). The gains depend on model capability: models without native tool calling, reasoning models, and smaller models gain substantially (+24.0pp to +43.1pp), while heavily optimized frontier models (GPT-5, Gemini 2.5 Pro) show smaller or reversed advantages. This matches recent analyses of reinforcement-learning-optimized tool use (Martinez, 2025). NLT also cuts token usage by 25.2%. The reliability and efficiency advantages compound in recursive agentic workflows, where agents chain many tool calls across sub-agents: a structured failure triggers retries, fallback routing, and coordination overhead, while NLT avoids most of that cost at the source. This work makes three contributions: (1) the first independent validation of NLT using open-source tooling, (2) evidence that model capability moderates NLT's advantages (Chen et al., 2025; Zhang et al., 2025), and (3) a measurement of NLT's reliability benefit (93% fewer errors), its most deployment-relevant property given the known fragility of structured tool calling. NLT is a practical alternative to structured tool calling, especially for production systems that value reliability over parseability.

Reinforcement Learning (RL) has emerged as a powerful paradigm for adaptive traffic signal control. However, in safety-critical infrastructure like traffic control, the opaque, black-box nature of deep RL models poses challenges for transportation agency acceptance, regulatory compliance, operational trust, troubleshooting, and fine-tuning. To bridge this gap between high-performance optimization and human-comprehensible interpretability, this effort introduces a novel, explainable entity centric RL framework for safe and transparent traffic signal control. Rather than processing traffic states through monolithic, flat vectors, the proposed architecture disaggregates real-time intersection observations into distinct, high-dimensional lane entities and phase temporal configurations to inherently preserve the structural topology and geometric configurations of the intersection. Relational dependencies and inter-lane conflicts are dynamically extracted via a dual-stage attention network featuring sequential multi-head cross-attention and self-attention blocks. This design yields a real time affinity matrix that quantifies the direct influence of signal phases on specific approach volumes and queues, providing full visual and analytical interpretability. To ensure strict operational reliability, a deterministic action-masking interface is integrated directly into the Proximal Policy Optimization pipeline, explicitly blocking invalid phase transitions to guarantee absolute compliance with established signal timing and safety constraints. Evaluated in a microscopic simulation environment, outperforms state-of-the-art baselines in delay minimization. More importantly, the emergent attention weights align precisely with established traffic engineering principles, offering an auditable, trust-enabling, and deployable architecture for next-generation adaptive traffic control systems.

Large language model (LLM) agents have shown strong decision-making capabilities in long-horizon interactive tasks, yet they still struggle to effectively leverage failed trajectories: full retries incur high interaction costs, while experience retrieval tends to dilute critical experience signals. To address this, we propose PivoARL, a self-feedback retry framework for experience exploitation in LLM agents. PivoARL identifies the pivotal erroneous turn through structured reflection and performs local retry only from the corresponding pivotal state, thereby reusing the correct prefix and reducing redundant interactions. From an information-gain perspective, we further show that pivotal retry concentrates useful experience signals near the error boundary, mitigating the signal dilution caused by state-agnostic experience utilization. Based on this insight, we design a pivotal-aware credit assignment mechanism that rewards correct prefixes while isolating erroneous suffixes, and optimize reflection quality through implicit reflection returns. We conduct a systematic evaluation on 4 agent tasks and 7 search-based QA benchmarks. Results show that PivoARL achieves significant improvements on Pass@2/3 across all tasks, with an average gain of about 11.5% over MetaRL. Moreover, benefiting from contrastive preference signals induced by pivotal turns, PivoARL also consistently improves Pass@1 on over 80% of the tasks. On Minesweeper environment, PivoARL improves over GiGPO by more than 45% and reduces interaction turns by about 42% on average compared with full-retry methods. Code is available at https://github.com/yuki-younai/PivoARL.

Deep reinforcement learning (DRL) in industrial control often suffers from lag and overshoot due to purely reactive control based on the current tracking error. To achieve anticipatory control without high computational overhead, we introduce a predictive formulation that augments the DRL state space with target velocities and future reference horizons. Evaluating eight configurations using proximal policy optimization (PPO) on a 1-degree-of-freedom (1-DoF) helicopter testbed, simulation results showed a 9-fold error reduction, lowering the mean absolute deviation from 2.73° to 0.31°. However, zero-shot transfer to physical hardware revealed a sim-to-real gap. Interestingly, a simpler configuration using a single, further look-ahead horizon matched the real-world top performance of the most complex model (1.11°). Overall, evaluating various combinations of prediction horizons and target velocities demonstrated that highly granular predictive data is not necessarily required for physical transfer.

Deep reinforcement learning (DRL) offers powerful control for industrial cyber-physical systems (ICPSs), but its "black-box" exploration risks violating strict hardware safety limits. Typically, these constraints are managed through complex reward shaping. In this work-in-progress paper, we embed a differentiable physics model directly into the proximal policy optimization (PPO) actor loss function. By simulating short-horizon future trajectories during training, the policy is penalized for anticipated safety violations independent of the task-reward signal. Evaluated on a simulated 1-degree-of-freedom helicopter testbed with strict pitch constraints, our physics-informed soft regularizations substantially reduce constraint violations while maintaining reliable target tracking.

End-to-end autonomous driving (E2E-AD) aims to directly map raw sensor information to driving actions. Recently, with the rapid advancement of multi-modal large language models (MLLMs), researchers have proposed the paradigm of Vision-Language-Action (VLA) models for E2E-AD, where it seeks to integrate visual perception, language understanding and action prediction within a single policy. However, existing VLA-based policies largely adopts imitation learning, where it only learns to drive by optimizing distance-based metrics w.r.t. logged expert trajectories. Such distribution shift between open-loop training and closed-loop inference leads to suboptimal performance in closed-loop planning. To close this gap, we present CLEAR, a system that enables closed-loop training using Reinforcement Learning (RL) at scale for E2E-AD. We propose to learn a novel residual waypoint policy around the waypoint prior from pretrained VLA policies, effectively harnessing the knowledge within. On another front, one of the key challenges to scale up RL for vision-based policies is the number of parallel simulation environments since RL is data hungry. To that end, we design a heterogeneous pipeline that places the simulator and the VLA learner on distinct compute groups, which allows us to dramatically increase the number of simulation environments running in parallel while avoiding resource contention and maintaining training stability. We show that with a simple reward, CLEAR significantly outperforms previous methods and sets new state-of-the-art performance on the challenging benchmarks of CARLA longest6 v2 and Bench2Drive.

Large vision-language models can reason over multimodal inputs by generating textual chains of thought (CoT). A key capability exhibited in CoT reasoning is self-reflection: revisiting earlier decisions and correcting previous errors. However, existing LVLMs often fail to properly attend to visual inputs during reflection, limiting their ability to translate feedback into grounded corrections, especially for out-of-distribution images. To address this issue, we propose a novel reinforcement learning training framework VRRL, with two components explicitly designed to elicit visually grounded self-reflection. First, we randomly mask trajectory prefixes during training to emphasize recovery from incorrect intermediate predictions rather than making early mistakes. Second, we introduce buffered roll-ins from an experience replay buffer to expose the model to diverse failure states that it must learn to correct. We evaluate our approach on visual grounding tasks involving tables and charts, as well as spatial navigation benchmarks. While off-the-shelf and conventionally fine-tuned models degrade substantially under distribution shift, our method substantially improves average out-of-distribution accuracy over standard RL and reflection-oriented fine-tuning baselines by using self-reflection effectively.

This paper presents a methodology for learning a control policy to intercept an intruder using the 3D direction unit vector to the intruder and the interceptor state. Prior deep reinforcement learning approaches assume either relative position or distance to the intruder is available, but this information is not readily accessible in real-world applications that employ passive, monocular camera sensors. Instead, we propose a solution that leverages an analytical policy gradient method using differentiable quadrotor dynamics to learn agile interception at speeds up to 10 m/s. The proposed approach outperforms baseline methods that utilize simplified point mass dynamics by an average of 30%.

Most LP-from-text benchmarks are static datasets of word problems written and labeled by hand. Once such a dataset is released, its size is fixed, its difficulty is fixed, and every problem can leak into the training data of future LLMs. We present \textbf{A$^{2}$utoLPBench}, a benchmark for testing LLM-driven agents on linear programming problems written in plain text. We first pick a feasible point and dual, then write down a problem for which that point is optimal and the objective value is known. The answer is known by construction, with no solver call and no human annotator. The evaluation environment bundles a reference solver-critic baseline and a Docker image whose usage instructions are written for an LLM-driven agent to read. With these in place, any agent can run the benchmark and get a calibrated score with one command. Because the benchmark is a generator rather than a fixed dataset, it has properties no fixed dataset can match: an unlimited supply of fresh problems, a difficulty knob set by $(n,m)$, ground-truth answers correct by construction, low LLM-side cost per problem relative to human authoring, repeatable scores across independent batches, and resistance to training-data leakage when fresh post-cutoff seed ranges are used.

We study timestep allocation for score-based diffusion sampling, where a learned reverse-time dynamics is discretized on a finite grid. Uniform and hand-crafted schedules are standard choices, but they rely on fixed prescriptions and can therefore be suboptimal. To address this limitation, we propose Adaptive Reparameterized Time (ART), a continuous-time control formulation that learns a time change by treating the speed of the sampling clock as the control, so that a uniform grid on the learned clock induces adaptive timesteps in the original diffusion time. Based on a leading-order Euler error surrogate, ART provides a principled objective for allocating timesteps along the sampling trajectory. To solve this deterministic control problem, we introduce ART-RL, an auxiliary randomized formulation with Gaussian policies that turns schedule learning into a continuous-time reinforcement learning problem. We prove that the randomized ART-RL formulation is equivalent to ART at the optimizer level, in the sense that its optimal Gaussian policy recovers the optimal ART time-warping rate through its mean. We further establish policy evaluation and policy improvement characterizations and derive trajectory-based moment identities that yield implementable actor--critic updates for learning the schedule. Across experiments ranging from controlled low-dimensional settings to image generation, ART-RL can be plugged into existing diffusion samplers by changing only the timestep grid, consistently improving sample quality over strong baseline schedules at matched budgets while leaving the rest of the sampling pipeline unchanged. The learned schedules also exhibit broad generalization, transferring without retraining across sampling budgets, datasets, solvers, pipelines, and representation spaces.

Long-context reasoning requires models to locate, revise, and synthesize evidence distributed across lengthy inputs. Existing long-context RL methods usually reward final answers or static evidence extraction, offering little feedback on how intermediate actions change the model's evidence state. We propose Maven, a reinforcement learning framework with an editable evidence memory. Maven defines an answer-conditioned evidence-state value and rewards action-level state transitions: add actions are credited by marginal gain and hindsight contribution, link actions by evidence synergy, and drop actions by improved answer support after removing misleading evidence. These rewards are assigned to the corresponding action spans in GRPO. Across Llama and Qwen models on LongBench v2, LongReason, and RULER, Maven outperforms outcome-only RL and evidence-identification baselines, producing more sufficient evidence sets and lower distractor retention. Our results show that long-context RL benefits from optimizing stateful evidence navigation rather than one-shot evidence extraction.

Value functions are an essential component in actor-critic based deep reinforcement learning (RL). Conventionally, these functions are trained as a regression task by minimising the mean squared error (MSE) relative to bootstrapped target values. Meanwhile, in distributional RL, a distribution of returns is modelled based on the distributional Bellman operator. This work investigates the Gaussian Histogram Loss (HL-Gauss), a recent approach that reframes value estimation as classification by encoding each scalar Bellman target as a Gaussian-smoothed categorical target. Despite its potential, applying histogram-based losses to RL presents inherent challenges, most notably the requirement to pre-define a fixed support interval, which is often complicated by the non-stationary and stochastic nature of target values typically found in RL tasks. In this work, we propose an approach that dynamically learns the lower and upper bounds of the support instead of assigning them beforehand. We derive an objective that jointly learns these bounds whilst learning the categorical representation of the scalar values, and we show that this objective forms an upper bound on the mean-squared Bellman error. Our theoretical analysis further shows that this bound is tighter than that of non-learned supports of HL-Gauss. Empirically, the proposed objective enables stable adaptation of the support interval and matches HL-Gauss-based actor-critic algorithms on most continuous-control tasks whilst improving on a subset, without requiring a pre-specified support interval.

With the rapid development of autonomous aerial systems, Unmanned Aerial Vehicles (UAVs) are increasingly deployed in applications such as inspection, environmental monitoring, and rescue, creating growing demand for reliable autonomous navigation. However, autonomous UAV navigation in dense environments remains challenging under sparse perception and dynamic constraints. Most reinforcement learning (RL) methods lack explicit safety mechanisms, leading to unsafe exploration, unstable training, and risky behaviors, especially during high-speed flight. Even in safe RL approaches, safety is often enforced by projecting policy outputs onto a safe action set, which may introduce instability. Meanwhile, many learning-based methods rely on dense inputs or large networks, increasing computational burden and limiting lightweight onboard deployment. Facing the above challenges, we propose a safety-constrained perception-control integrated framework for UAV navigation. A lightweight network encodes sparse observations into collision-risk-aware features using asymmetric and depthwise separable convolutions. We formulate the task as a constrained Markov decision process within a hierarchical control architecture and solve it using a Lagrangian-based safe PPO algorithm. Curriculum learning further improves training stability. Experiments with varying obstacle densities and flight speeds demonstrate higher success rates, improved safety, and better efficiency than existing reinforcement learning baselines.

This monograph provides an introduction to mean field reinforcement learning through the lens of Markov decision processes arising from large-population stochastic control with mean field interactions and common noise. Starting from the connection between multi-agent reinforcement learning and mean field control, it develops the probabilistic, mathematical, and control-theoretic framework needed to formulate representative-agent learning problems, analyze their relationship with finite-population systems, and study both general and linear-quadratic models. The presentation includes dynamic programming principles, propagation-of-chaos limits, and theoretical analyses of tabular Q-learning and policy-gradient methods. It also discusses numerical implementations, including tabular schemes and deep reinforcement learning methods such as deep deterministic policy gradient. The goal is to give readers a coherent bridge between mean field control theory and reinforcement learning methodology, emphasizing the mathematical structure of the problems and the design of tractable learning approaches for large stochastic populations.

Reinforcement learning (RL) has become a central component of post-training large language models (LLMs), yet little is understood about how RL adaptation is distributed across transformer layers. Existing approaches typically update all model parameters uniformly, implicitly assuming that every layer contributes similarly to the gains obtained during RL post-training. In this work, we challenge this assumption through a systematic layer-wise study of RL training. Surprisingly, we find that training a single transformer layer can recover most of the gains achieved by full-parameter RL training, and in some cases even surpass it. To quantify this phenomenon, we introduce the quantity layer contribution, which measures the fraction of full RL improvement recovered by training a layer in isolation. Across seven models spanning two model families (Qwen3, Qwen2.5), three RL algorithms (GRPO, GiGPO, Dr. GRPO), and multiple task domains including mathematical reasoning, code generation, and agentic decision-making, we observe a remarkably stable pattern: RL gains are highly concentrated in a small subset of, and in many cases even a single, transformer layers. More strikingly, the same structural pattern consistently emerges: high-contribution layers concentrate in the middle of the transformer stack, while layers near the input and output ends contribute substantially less. The resulting layer rankings remain strongly correlated across datasets, tasks, model families, and RL algorithms.

Collaborative capture of dynamic targets is common in nature as an essential strategy for weaker species against the strong. Similar concepts have shown to be useful for numerous robotic applications, such as security and surveillance, search and rescue. However, most existing works focus on analytical and geometric solutions or end-to-end reinforcement learning methods, which are largely constrained to obstacle-free environments or scenarios with sparse, regularly distributed obstacles. This work tackles the problem from a unique perspective: the renowned strategy of``ambush'' alone would suffice for multiple slower pursuers to capture one faster evader with different levels of intelligence efficiently in complex environments. A parameterized strategy of ambush (including discrete and continuous parameters) is designed first, which takes into account the topological properties of the workspace, the truncated line-of-sight visibility, the relative speed ratio and the limited capture range. Then, a Hybrid Monte Carlo Tree Search (H-MCTS) algorithm is proposed to optimize the associated parameters through long-term planning, enabling the identification of highly promising parameters for future capture. Lastly, the neural acceleration is trained offline to learn the ranking of different choices of parameters across various environments, and to directly predict scores, replacing the rollout process in H-MCTS. The neural acceleration is adopted during online H-MCTS to accelerate the planning procedure while guaranteeing the planning quality. Its efficiency and effectiveness are validated in extensive simulations and hardware experiments, against evaders with different capabilities and intelligence levels, including two-times higher velocity and human-controlled behavior.

Accelerating materials discovery requires AI systems that can generate scientifically valid hypotheses through multi-step, domain-grounded reasoning. Standard large language models often produce fluent but weakly traceable responses to open-ended materials design problems, making it difficult to determine whether final answers are supported by coherent intermediate reasoning. We develop Graph-PRefLexOR, a family of graph-native reasoning models fine-tuned with Group Relative Policy Optimization (GRPO) to organize reasoning into explicit phases for mechanism exploration, graph construction, pattern extraction, and hypothesis synthesis. This design links neural language generation with symbolic relational structure, enabling causal connections to be constructed, inspected, and reused. On 100 open-ended questions from materials science and mechanics literature, Graph-PRefLexOR achieves 40-65% improvements over corresponding base models, with the largest gains in reasoning traceability. Embedding analyses show broader semantic exploration and approximately 2-3 times greater semantic diversity than baselines. Semantic backtracking and layer-wise hidden-state analyses further show stronger alignment between structured reasoning and final answers. Finally, test-time graph expansion reveals that additional compute primarily increases long-range conceptual recombination within a bounded semantic space, rather than simply expanding semantic coverage. These results establish graph-native reinforcement learning as a pathway toward interpretable AI systems for scientific hypothesis generation in materials design and other scientific applications.

The development of decision-pretrained transformers, algorithm distillation, long-context meta-RL, and retrieval-augmented agents has renewed interest in in-context reinforcement learning (ICRL): the ability of a pretrained or fine-tuned decision model to infer latent task rules and improve future behavior from interaction context, without test-time parameter updates. This line of work asks when trial-and-error evidence, rewards, transitions, demonstrations, feedback, or retrieved experience can make learning-like computation happen inside the context window. However, existing surveys of ICRL mainly organize the field around pretraining objectives, architectures, context formats, evaluation protocols, and theoretical mechanisms, while the non-stationary setting remains comparatively underexamined. In changing environments, accumulated context is not merely more evidence about a fixed task: the reward specification, transition kernel, observation channel, action interface, constraint model, or demonstration and memory distribution can fall out of alignment with the current regime. Previously useful context can therefore become stale, misleading, or useful again when an old regime returns. We survey non-stationary ICRL as the problem of adapting through context while deployed policy parameters remain fixed: the policy must infer both the current decision rule and which parts of its accumulated evidence still support that rule. We define non-stationary ICRL, relate it to meta-RL, decision sequence modeling, retrieval-augmented RL, value- and model-aware ICRL, and reward-feedback agents, and organize the literature along three questions: what changes, how the change unfolds, and how observable the change is to the agent.

Visual Reinforcement Learning (VRL) has achieved considerable success in solving control tasks. However, generalizing learned policies to new environments remains a major challenge, as agents often overfit to task-irrelevant features in the training environment. To solve this problem, we introduce the concept of decoupling observations into task-relevant and task-irrelevant representations. Building on this idea, we propose a self-supervised Task-Relevant Representation Decoupling (T2RD) algorithm for VRL. This algorithm consists of three components: task-relevant representation consistency, cross-reconstruction, and cross-dynamic prediction. The first two components achieve the decoupling of content and style features, but the resulting content representations are not necessarily task-relevant. To further refine task-relevant features from content representations, we design the third component that introduces dynamic prediction. T2RD achieves State-Of-The-Art (SOTA) generalization performance and sample efficiency in the DeepMind Control Suite and Robotic Manipulation tasks.

Generalist robot policies learn a diverse repertoire of behaviors from large-scale pretraining. In principle, this makes them excellent priors for downstream adaptation via reinforcement learning (RL). In practice, however, standard RL methods leveraging this prior optimize directly over robot actions, requiring the base policy's action distribution to be close to that of a performant policy from the start. This assumption breaks down for complex or long-horizon tasks that fall outside the pretraining distribution. Our key insight is that, for sufficiently expressive generalist policies, language prompts are an effective alternative space for learning to solve such tasks: modulating language inputs elicits skills already within the policy's repertoire, which can be composed to solve tasks beyond its zero-shot capabilities. We propose Semantic Action Reinforcement Learning (SARL), which learns to optimize this prompt space through online interaction, treating the generalist policy as a controllable skill prior. Importantly, leveraging pretrained skills rather than learning new ones from scratch yields structured, semantically meaningful exploration and highly efficient online improvement, and learning to modulate prompts through experience grounds them in induced real-world behaviors for robust task-solving. Across real-world settings and simulated benchmarks, we show SARL unlocks fundamentally new capabilities -- adapting VLA behavior to solve complex, long-horizon tasks -- and significantly outperforms existing approaches for improving robot behavior in deployment.

Vision-Language-Action (VLA) models offer a promising framework for robotic manipulation by connecting language instructions, visual observations, and continuous control. However, most existing policies remain limited by behavior cloning or supervised fine-tuning (SFT) from fixed demonstrations, which provides limited opportunity to improve from the policy's own failures. In this paper, we present Z-1, a reinforcement learning (RL) post-training framework for flow-based VLA models. Built on top of $π_{0.5}$, Z-1 uses only publicly released RoboCasa demonstrations for SFT and then applies a task-wise Group Relative Policy Optimization (GRPO) strategy across $24$ standard RoboCasa tasks. To improve the efficiency and stability of online optimization, Z-1 combines shared-prefix rollout construction, tree-structured trajectory branching, completion-aware reward calibration, and selective joint training of VLM and Action Expert. Across all $24$ RoboCasa tasks, Z-1 achieves an average success rate of $80.6\%$, improving over its SFT initialization by $13.2\%$ points and outperforms the published sota models. These results show that systematic GRPO post-training can substantially improve flow-based VLA policies without additional private demonstrations.

Recent multimodal large language models have shown great promise in clinical image reasoning, but existing post-training pipelines remain predominantly outcome-centric, relying on final answer correctness or sequence-level preferences. This suffers from sparse credit assignment, making it difficult to optimize the reasoning process essential for clinical applications. Our analysis reveals that cascading errors from early-stage reasoning failures are a leading cause of incorrect predictions in medical visual question answering (VQA) benchmarks. Motivated by this, we propose Medical Reasoning-aware Policy Optimization (MRPO), an RL algorithm that incorporates step-wise process rewards. When the final answer is incorrect, MRPO assigns exponentially larger penalties to tokens in earlier invalid reasoning steps, breaking failure cascades without compromising successful paths. Across three multimodal LLM backbones, MRPO consistently outperforms standard GRPO and a recent RL baseline, and on Qwen3-VL-8B-Instruct even surpasses substantially larger medical MLLMs such as HuatuoGPT-Vision-34B by 2.79 points. Moreover, MRPO reduces early-stage reasoning failures from 64.0% to 13.0%, showing that targeted mitigation of cascading failures improves both reasoning quality and final answer accuracy. Our code is available at https://github.com/dmis-lab/MRPO

Scalable reinforcement learning has popularized high-throughput sampling architectures, which significantly compresses the training time for off-policy methods in robotic locomotion. However, the rapid increase of data volume and update frequency undermines the stability of value-based methods and diminishes the plasticity of policy networks. To address these challenges, this work presents FastDSAC, a fast and high-performance variant of the Distributional Actor-Critic algorithm designed for parallel sampling scenarios. Specifically, we introduce a truncated Gaussian distribution to approximate the learned policy, which effectively excludes out-of-distribution actions that strain target value estimation while keeping necessary stochasticity for exploration. The proposed action constraint functions as an implicit regularization, which counteracts the plasticity loss typically caused by aggressive gradient updates. This preservation of network adaptability enhances sample efficiency, particularly in scenarios with a high update-to-data ratio, and accelerates the early training process. In contrast to prior fast reinforcement learning approaches that rely on discrete value distributions, our method utilizes a continuous Gaussian representation equipped with adaptive variance regulation, which improves value estimation accuracy by sampling confident and informative transitions. Extensive experiments on MuJoCo Playground and HumanoidBench demonstrate that FastDSAC not only stabilizes the overall training process but also achieves superior asymptotic performance and faster convergence compared to state-of-the-art baselines.

Long-horizon language agents must repeatedly interact with tools, accumulate evidence, and make decisions under bounded context windows. Context-management methods make such rollouts feasible by simplifying past interactions through deletion, folding, or memory editing. However, when useful history is collapsed into compressed states, the reconstructed context may no longer reveal which earlier observations support a successful final answer. This creates a mismatch between bounded-context acting and outcome-based reinforcement learning: the policy acts on reconstructed context, while the learner lacks source-level provenance for assigning credit to the evidence that mattered. We propose ECHO, a selective turn-memory framework for traceable context reconstruction in Agentic RL. ECHO compresses each completed environment turn into a compact source-indexed memory record, reconstructs bounded policy contexts by selecting useful records, and reuses the selected source indices to route positive outcome credit to the final trajectory segment, reused evidence turns, memory findings, and memory-selection actions. On BrowseComp-Plus, ECHO reaches 43.4% held-out accuracy, outperforming GRPO at 28.9% and the rolling-summary baseline SUPO at 36.1%, while using fewer turns and lower trajectory volume than SUPO. The trained policy also improves zero-shot generalization across multi-objective QA, code generation, and deep information-seeking benchmarks on both dense and MoE backbones.

Autonomous aerial inspection of marine infrastructure is frequently compromised by stochastic sea states, introducing risks of high-kinetic impacts, post-landing toppling, and sensory occlusion. This paper proposes a decoupled, multi-vehicle landing framework synchronizing an Unmanned Surface Vehicle (USV) equipped with a 3-RPU stabilized platform with a robust Unmanned Aerial Vehicle (UAV). The architecture utilizes two independent Deep Reinforcement Learning (DRL) agents: a Soft Actor-Critic (SAC) agent providing high-frequency wave-motion compensation for the landing deck, and a multimodal RL agent for the UAVs final approach. Evaluated in high-fidelity maritime simulations, the system achieved a 100% landing success rate across 15 trials in wave states varying from calm to rough. Results show a mean stabilization efficacy of 87.8%, maintaining the landing surface within 1 degree of the horizontal plane for 96% of the mission duration in rough conditions, effectively contributing to safer landings.

Safe online reinforcement learning requires policies to respect safety constraints while maintaining smooth optimization dynamics. Existing approaches typically rely on either strict safety enforcement via action interventions, which introduce discontinuities in system interaction and learning, or soft safety constraint formulations, which preserve smooth learning but provide limited safety assurance. We propose AutoSafe, a safety-aware policy architecture that integrates structured safety monitoring and intervention directly into the action generation process. This design enables smooth, risk-dependent transitions between performance-driven and safety-preserving behaviors, resulting in continuous online interaction and learning dynamics. Empirical results across a suite of continuous-control benchmarks demonstrate strong safety enforcement without sacrificing learning smoothness. We further validate AutoSafe on a physical cart-pole system, highlighting its practical effectiveness for safe online learning in the real world.

Deep reinforcement learning has the potential to solve attitude control problems more adaptively, precisely, and robustly by handling nonlinear dynamics, uncertainties, and failure cases more effectively than traditional attitude control approaches. We explore reinforcement learning (RL) for attitude control in spacecraft re-entry. An industry-standard proportional-integral-derivative controller with gain scheduling serves as a strong baseline for model-free RL and hybrid controllers that combine these two approaches. We formalize the application in the RL framework to apply continuous, off-policy RL. State-of-the-art RL achieves comparable performance to traditional control approaches in this domain. However, its out-of-distribution generalization is not sufficient. Hence, we use dynamics randomization to introduce challenging task variations during training and enforce generalization in a predefined operational envelope. Finally, we assess the best obtained RL-based controllers with application-specific metrics to show superior performance in comparison to traditional controllers in the operational envelope, that is, hybrid controllers are able to track the angle of attack better and are more robust under variations of mass, inertia tensor, and flap actuator bandwidth.

Embodied task planning asks an agent to turn a natural-language instruction into an executable sequence of actions in a physical scene, and is a building block for household, assistive, and service robots. Recent prompting-based and reinforcement-learning planners generate fluent action text but lack a cheap deterministic check that the produced plan is valid in the target world, while high-fidelity simulation is too slow to serve as an inner-loop training signal. The general problem is therefore how to obtain verifiable supervision and rewards for embodied planners without relying on string-level matching or full simulation. Here we show that a single BDDL specification, automatically constructed from open-world video evidence or curated tasks, can serve as a shared interface for data construction, plan verification, and reward design. A video-to-BDDL parser, an LLM verifier, and a lightweight symbolic engine together supply dense feedback at millisecond latency. We further introduce GroupAdapt, a difficulty-aware length schedule that uses the in-batch group pass rate as a zero-cost signal so that hard prompts get wider length tolerance and automatically tighten as their pass rate improves. Under the guidance of the proposed verifier and GroupAdapt schedule, the 8B planner attains a Strict-Pass score of 97.3 on BEHAVIOR-1000, yielding a 25.9 percent relative improvement over the Qwen3-8B baseline. This result exceeds the strongest large-model baseline by 3.5 percent, while simultaneously compressing the response length by 79 percent to 207 tokens, demonstrating both effectiveness and efficiency.

Residual reinforcement learning adapts a pretrained robot policy by learning an additive correction to its actions. While effective when adaptation amounts to shifting the base policy's action distribution, additive corrections cannot change the distribution's shape, scale, or state-dependent geometry -- limitations we formalize as wrong variance, miscalibrated confidence, and non-uniform correction. We show that these matter under dynamics shift: when the base distribution is geometrically mismatched to the shifted system, residual correction can underperform even the unadapted policy. We propose Warp RL, a policy adaptation method that replaces additive residuals with an invertible, state-conditioned transformation of the base policy's action distribution. Instantiated with monotonic rational-quadratic spline flows (arXiv:1906.04032), Warp RL preserves identity initialization, strictly generalizes additive residual correction, and exposes a structured adaptation space suitable for both policy-gradient and gradient-free optimization. Across a variety of ManiSkill3 manipulation tasks with controlled dynamics shifts, Warp RL matches residual correction when translation is sufficient and substantially outperforms it when adaptation requires distributional reshaping. We further demonstrate that warping can replace additive correction in an off-policy sim-to-real pipeline, achieving comparable success rate with 30% faster task completion on a real-robot peg-insertion task.

Active flow control is a fundamental application in engineering. Recent advances in deep reinforcement learning have made progress in this field. However, the classical online RL approaches require extensive real-time interactions with the high fidelity environment, while each sensor configuration change necessitates whole policy retraining. All these factors result in prohibitive computational costs for real-world applications. In this work, we propose a novel offline RL framework that addresses both challenges through data-driven policy extraction. We develop a sensor position-conditioned architecture that enables a single policy network to adapt seamlessly to multiple sensor arrangements. The position-conditioned approach incorporated spatial relationship modeling through Point Attention layers to ensure the generalizability to varying sensor placements. We demonstrate the framework on two representative problems, mitigating chaoticity in the Kuramoto-Sivashinsky equation and flow control over airfoils governed by the Navier-Stokes equation. The result demonstrates that the policy extraction from the dataset provides unprecedented flexibility for sensor placement optimization. This approach represents a significant step towards adaptive, intelligent flow control systems.

We present a three-phase deep reinforcement learning system for personalized portfolio management that addresses three limitations shared by all prior financial RL work: 1) ticker lock-in, 2) monolithic objectives , and 3) static user models. Phase 1 pretrains a ticker-identity-free cross asset encoder via self-supervised learning on a multi-asset corpus, augmented by a frozen parallel branch using Chronos, a T5-based time series foundation model, fused via a learned gating mechanism. To our knowledge, this is the first application of a time series foundation model to portfolio management RL. The encoder generalizes to any publicly traded asset via a 50-dimensional observable metadata vector that requires no retraining for new tickers. Phase 2 fine-tunes a MoE (Mixture of Experts) portfolio actor critic with PPO under an objective-conditioned reward that simultaneously serves six distinct investment goals sampled per episode: short-term alpha, short-term gain, long-term gain, capital preservation, tax-loss harvesting, and long-term-gains-only. A MoE architecture assigns each objective to a specialized expert head (momentum, growth, defensive, tax-aware), and a learned intent router blends experts based on the active objective and current market regime, which eliminates cross-objective gradient conflict. Phase 3 adds a lightweight personalization layer further adapted at inference time to each individual via a 76-parameter LoRA module fine-tuned on real brokerage transaction history, inferring investment objectives from revealed trading behavior rather than questionnaires. A natural language intent parser converts free-form goals directly into structured investment objective parameters.

Micro-ultrasound ($μ$US) is a new, emerging, and promising imaging modality for prostate cancer (PCa) detection, but accurate identification of suspicious tissue remains highly dependent on clinical experience, leading to substantial inter-observer variability. Machine-learning assistance can reduce this variability; however, training reliable deep models is challenging because supervision is sparse and noisy -- typically limited to core-level histopathology outcomes (e.g., cancer grade and its percentage in a biopsy core) without pixel-level lesion annotations and under severe class imbalance. We introduce Prost-RL, which reframes $μ$US PCa detection as a spatially aware, policy-driven inference problem by learning where to look before decoding. Prost-RL integrates a lightweight reinforcement-learning policy into a foundation-model encoder-decoder to generate interpretable spatial attention maps that act as soft prompts for both cancer-likelihood heatmap prediction and image-level classification. We further propose Adaptive Policy Optimization (APO) to stabilize hybrid supervised-RL training and a noise-robust objective combining symmetric cross-entropy with negative-entropy regularization to mitigate weak-label noise and encourage sharp localization. On a cohort of 6,607 biopsy cores from 693 patients across five clinical sites, Prost-RL achieves $79.0\pm3.5$ AUROC with $64.6\pm6.3$% sensitivity at 80% specificity for core-level detection (+2.1 AUROC and +4.5 sensitivity points over the strongest baseline), and $79.3\pm5.8$ AUROC for clinically significant cancer classification. The learned policy highlights biopsy-aligned regions, providing transparent, spatially grounded evidence alongside quantitative risk predictions. Code is available at: https://github.com/DeepRCL/Prost-RL.

Modern large language models (LLMs) rely on reinforcement learning during post-training to push specific capabilities, yet integrating multiple capabilities into one model remains hard. Existing methods, such as Off-Policy Finetune and Mix-RL, are either inefficient or lose performance. In this work, we propose Multi-teacher On-Policy Distillation (MOPD), a post-training paradigm for combining the capabilities of multiple domain RL teachers: we first run per-domain specialised RL to obtain a set of domain teachers, then distill these teachers into the student on its own rollouts. This eliminates exposure bias and provides a dense optimization signal. On Qwen3-30B-A3B, MOPD outperforms Mix-RL, Cascade RL, Off-Policy Finetune, and Param-Merge baselines, inheriting nearly all of each teacher's capability. MOPD also enables parallel, independent development of domain teachers, removing the cross-domain coupling typical of multi-domain post-training. MOPD has been deployed in the post-training of MiMo-V2-Flash, an industrial-scale frontier model, demonstrating its practical value for capability integration in frontier-scale LLMs.

Improving vision-language models (VLMs) on visual reasoning typically requires retraining or hand-designed prompts and tools. We present Dynamo, a training-free framework that adapts a frozen VLM without any weight updates. On a small labeled training subset, the agent inspects its own correct and incorrect attempts and evolves two complementary capabilities: reusable reasoning skills for cognitive bottlenecks, and executable visual tools for perceptual ones. Each generated tool is paired with a skill that specifies when to invoke it, and both capability types accumulate in a persistent library. Across four visual reasoning benchmarks and five VLM backbones, Dynamo improves direct inference on all 20 model--benchmark settings (avg. +5.6 acc). When the tool set is given in advance, the framework learns when to call each tool, and per-step tool choice improves on every tested backbone. Against task-specific RL (VTool-R1, DeepEyes), Dynamo closes 65--99% of the RL gap at a fraction of the compute, and combines additively with RL when available.

Real-time strategy (RTS) games present significant AI challenges, characterized by expansive state-action spaces arising from multi-unit coordination in continuous battlefields, and sparse delayed rewards stemming from final win/lose signals. Existing approaches face a trade-off between managing the dimensionality explosion of joint actions and maintaining the interpretability of complex state representations. This complexity is further intensified by the limitation of traditional hierarchical structures in adaptively decomposing tasks into effective tactical modules. Such difficulties are compounded by the black-box nature of deep learning models and their reliance on sparse rewards, which together result in limited sample efficiency and a lack of decision-making transparency. To address these limitations, this paper proposes HRL-IM/CBS, a hierarchical reinforcement learning framework with influence map hashing and cluster-based scripts for StarCraft micromanagement. Influence map hashing encodes global battlefield situations into compact hexadecimal codes, capturing spatial control and relative advantage. Cluster-based scripts enable dynamic local coordination through adaptive unit partitioning. The hierarchical multi-Q-table architecture decomposes decision-making into upper-level clustering strategy selection and lower-level tactical execution, with reward allocation providing dense learning signals. Experiments across six asymmetric scenarios demonstrate competitive performance against deep RL baselines while offering advantages in sample efficiency and interpretability through transparent Q-table representations.

Reinforcement learning (RL) has become indispensable for pushing Vision-Language-Action Models (VLAs) beyond static imitation learning. However, existing RL methods typically require external environmental feedback, relying on predefined success signals to guide policy updates. In this work, we show that VLA models possess useful internal evaluative capabilities: in discrete-action VLAs, trajectories with higher generation confidence are significantly more likely to succeed. Based on this observation, we introduce T^2VLA (Test-time VLA), an architecture-agnostic test-time RL framework that enables VLA models to achieve self-bootstrapping policy improvement. Instead of relying on external rewards, T^2VLA leverages trajectory-level similarity to high-confidence expert demonstrations as an intrinsic reward signal. In addition, we propose a Confidence-Driven Dual Expert Bootstrapping mechanism, which dynamically balances a Local Pseudo-Expert for exploration and a Global Expert Pool for training stability. Extensive experiments on the LIBERO and RoboTwin benchmarks show that T^2VLA consistently outperforms supervised baselines and approaches oracle RL performance with ground-truth rewards, achieving effective improvement without external reward feedback. Furthermore, T^2VLA adapts to distinct VLA paradigms, including both OpenVLA-OFT and the pi series.

Deep Reinforcement Learning (DRL) has achieved significant success in robotics and autonomous systems, yet remains vulnerable to adversarial perturbations that can severely degrade performance. Research in adversarial reinforcement learning is often limited by fragmented implementations, inconsistent evaluation protocols, and poor reproducibility. To address these challenges, we present \textbf{RoAd-RL}, an open-source benchmarking framework that provides unified abstractions for policies, attacks, defenses, and robustness metrics, together with reproducible evaluation pipelines and seamless integration with Stable-Baselines3 and Gymnasium. We evaluate DQN, PPO, and SAC agents in LunarLander and Highway-v0 under 192 attack-defense configurations. Results reveal substantial variations in robustness across environments and show that some commonly used defenses can be more detrimental than the attacks they aim to mitigate, while temporal smoothing consistently achieves strong performance. RoAd-RL establishes a standardized benchmark for adversarial reinforcement learning research and is publicly available at https://pypi.org/project/road-rl.

In complex continuous-control reinforcement learning tasks, multimodal optimal actions often coincide with uncertain, multimodal return distributions, making reliable value estimation and multimodal exploration challenging. Existing value estimation methods using unimodal Gaussians restrict expressiveness and yield biased estimates. Recent generative policies can represent multimodal actions but often collapse to a few modes and under-explore high-value areas of the action space. Motivated by these challenges, we propose Dual-Flow RL, a unified actor-critic framework that jointly models a continuous return distribution and a multimodal policy distribution using conditional flow matching (CFM). This design supports reliable value estimation and sustained multimodal exploration. To further enhance exploration, we introduce an Entropy-Covariance Exploration Regulator (ECER) that enables state-aware exploration regulation leveraging policy entropy and action-uncertainty covariance. Experiments on DeepMind Control Suite and Humanoid-Bench show that Dual-Flow RL achieves state-of-the-art performance on most tasks, significantly outperforming prior diffusion-based and flow-based methods.

Inductive biases steer learning toward generalizable solutions by encoding task structure. In this work, we identify a crucial missing bias in MLLMs: cross-view consistency, \textit{i.e.}, semantically invariant views of the same instance should lead to the same answer. Standard reinforcement learning with verifiable rewards (RLVR) objectives do not impose this constraint, but instead assign pointwise rewards to each visual input. Even with data augmentation (DA), transformed views are typically rewarded independently, providing little signal once within-view rewards saturate. We propose \textbf{ConsistRoll}, a simple but effective method that injects cross-view consistency into RLVR training by reusing the group-sampling mechanism of GRPO. Specifically, ConsistRoll places original and semantically invariant transformed views in the same generation group, and assigns a joint reward only when paired completions are both correct and consistent. In this way, ConsistRoll turns consistency into an online credit-assignment signal, \textbf{without extra generation overhead and annotations}. Theoretically, we show that cross-view consistency is a valid inductive bias, and ConsistRoll introduces a cross-view correction term absent from DA, penalizing view dependence and alleviating advantage collapse. Comprehensive benchmarks across math, general-purpose, hallucination domains confirm that ConsistRoll achieves robust improvements in multimodal reasoning.

Group Relative Policy Optimization (GRPO) is a default recipe for process-supervised reinforcement learning of LLM reasoners, and dense process supervision -- via learned process reward models (PRMs) or on-policy-distillation KL signals -- is a common way to densify its otherwise weak outcome reward. Layering such a step-level signal on top of GRPO's group-standardized advantage, however, exposes three structural pathologies: \emph{channel contamination} between the pooled process, outcome, and format streams at group standardization; \emph{resolution mismatch} between the granularity of the process signal and the granularity of the logical decisions being credited; and a \emph{cumulative trap} by which GRPO's return-to-go sum surfaces either length inflation or truncated exploration depending on the sign regime of the signal. We propose \textbf{PASS} (\emph{Process Advantage Signal Shaping}), a compact middleware that sits between any scalar step-level process signal and GRPO's clipped surrogate and addresses the three pathologies in turn: \emph{Advantage Fusion} standardizes the three streams independently within each group, \emph{Chunk-by-Value} derives value-homogeneous chunks from the signal itself and broadcasts credit within each chunk, and \emph{Divide-Length} converts the cumulative objective into an average-value-density score. We validate PASS across two domains and two process-signal paradigms -- a learned PRM on mathematical reasoning and an on-policy-distillation KL signal (with a generalized variant) on multi-hop question answering -- and under two group-standardization operators. In every regime PASS delivers a consistent pass@1 gain over the corresponding GRPO baseline.

Reinforcement learning with verifiable rewards (RLVR) has significantly improved the reasoning capability of large language models, reaching expert or even superhuman performance in domains such as competition math. However, whether weaker agents and humans can actually harness this capability is far less certain, with RLVR documented to drift reasoning toward idiosyncratic patterns such as poor readability and language mixing. Tandem training is a recently introduced paradigm that targets this compatibility problem: a trained, stronger senior co-generates each rollout with a frozen, weaker junior, and the two are rewarded as a team, so the senior is pushed to reason in ways the junior can follow. Yet this paradigm has so far been demonstrated only in proof-of-concept settings, leaving open whether it scales to the long chains of thought of the modern RLVR pipeline. In this work, we propose Tandem Reinforcement Learning (TRL), which carries the tandem training paradigm into RLVR. In TRL, the senior and a frozen junior alternate stochastically to co-generate the reasoning, the resulting generation is rewarded, and the standard GRPO loss is applied to the senior. Training Qwen3-4B-Instruct on competition math, we find that TRL matches vanilla GRPO on solo reasoning capability while three properties emerge together from the same rollout structure: stronger handoff robustness with the junior, reduced distributional drift from the junior, and a chain-of-thought more legible to the junior. Our results demonstrate a promising route for RLVR with practical payoffs in multi-model communication and human compatibility.

Geometry Problem Solving have increasingly adopt the neuro-symbolic paradigm, combining neural intuition with symbolic rigor. However, current frameworks suffer from severe bottlenecks in two core stages: autoformalization, which treats multimodal translation as a static task decoupled from downstream solver compatibility, and theorem prediction, where solvers frequently hit a deductive impasse due to fixed rule libraries. To address these, we propose SD-GPS, a solver-driven framework that treats the symbolic solver as an execution oracle throughout both formalization and deduction. First, Solver-Driven Autoformalization unifies supervised formal-language adaptation and solvability-guided reinforcement learning into a single module built on QwenVL3-2B, making executability the central training signal. Second, Verified Theorem Proposing introduces an impasse-aware agent that proposes local auxiliary lemmas from current proof states, ensuring soundness by filtering all proposals through symbolic verification. Empirical evaluations on Geometry3K and PGPS9K demonstrate that SD-GPS consistently outperforms existing MLLM, neural, and neuro-symbolic methods across standard completion, multiple-choice, and cross-modal reference regimes, proving that closing the loop between multimodal perception and symbolic execution significantly improves geometric reasoning, offering profound insights into how neural agents can be grounded by formal systems to achieve verifiable problem-solving capabilities.

One goal in reinforcement learning (RL) research is to understand general-purpose sequential decision-making, using benchmark simulators as a proxy for learning in deployment settings. When running experiments, however, the goal of achieving high performance in the simulator can mutate into focusing exclusively on solving the simulator. To achieve high scores, researchers may adopt solutions exclusively meant for solving simulators, rather than learning while the agent is deployed outside a simulator. Solving simulators is also worthy of investigation, but it is a fundamentally different RL research question. In this paper, we argue that RL researchers need to distinguish between two use cases of simulators: solving simulators and using simulators as a proxy for learning in deployment. We first discuss how these two use-cases are importantly different, in terms of constraints on how the agent can use the simulator, which algorithms are appropriate, and which evaluation metrics are appropriate. We then highlight several issues and misleading conclusions that can occur by not making the distinction between these two settings clear, supported with examples and simple experiments. This work is a call to the community to begin clearly distinguishing how they are using simulators in their work, hopefully sparking further discussion on which empirical practices work best in each setting.

We present Qwen-Image-2.0-RL, a post-training pipeline that applies reinforcement learning from human feedback (RLHF) and on-policy distillation (OPD) to improve both the visual quality and instruction-following capability of the Qwen-Image-2.0 diffusion model. To provide reliable reward signals, we construct task-specific composite reward models by fine-tuning vision-language models with a pointwise scoring paradigm and chain-of-thought reasoning. For text-to-image generation, the reward models cover alignment, aesthetics, and portrait fidelity dimensions. For image editing tasks, the reward system addresses instruction-following accuracy and face identity preservation. Building on this reward system, we develop a scalable GRPO-based RL training framework, incorporating a hybrid classifier-free guidance (CFG) strategy to preserve pre-trained knowledge, prompt curation via intra-group reward range filtering, and per-category reward weight calibration. To merge the task-specialized RL policies for T2I and editing, we propose on-policy distillation as the final training stage, which consolidates multiple teachers into a single student model through trajectory-level velocity matching. Extensive evaluation shows that Qwen-Image-2.0-RL achieves 57.84 overall score on Qwen-Image-Bench (+2.61 over the base model), Elo ratings of 1193 in text-to-image arena (+78) and 1349 in image edit arena (+93), demonstrating consistent gains in aesthetic quality, prompt adherence, and editing accuracy.

Quadrotors offer the agility needed to rapidly transport suspended payloads during time-critical applications, including search-and-rescue and medical delivery. While suspended-payload transport and traversal for these missions are well studied, the highly dynamic targeted release of the payload remains comparatively underexplored. State-of-the-art approaches typically rely on model-based trajectory optimization and tracking; however, these methods often yield sub-optimal performance due to conservative feasibility constraints, tracking errors, and the inherent difficulty of analytically modeling flexible rope dynamics. To overcome these limitations, we propose a hybrid simulation framework that couples a high-fidelity analytical quadrotor model with a physics solver for complex rope and payload interactions. By exchanging forces between the two domains at every step, we obtain a physically accurate simulation of the suspended-payload system. Leveraging this environment, we train a deep reinforcement learning (RL) policy that executes agile, accurate payload throws to designated targets. Deployed zero-shot on hardware, our RL policy pushes the boundary of the agility-accuracy trade-off, outperforming the model-based baseline by reducing the landing error by up to 50% and the throw duration by up to 30%. Ablation studies confirm that the coupled simulation is the key enabler of these gains. We further show that the same pipeline trains a policy driven by visual observations rather than an explicit state estimate, achieving accuracy comparable to that of the state-based policy. To accelerate future research in dynamic aerial manipulation, we open-source the simulator to the community upon acceptance.

Reinforcement learning from human feedback (RLHF) in production does not always have a synchronous reward signal. Code-execution verifiers, slow judge ensembles, and queued human review can return several gradient steps after the rollout that produced them, breaking the synchronous-reward assumption underlying standard PPO. We address this gap with Retroactive Advantage Correction (RAC): each pending slow completion is queued, aged through a non-negative kernel, and reinjected as a clipped residual into the next optimiser step's advantage. We prove that under an unbiased clipped importance ratio, the cumulative RAC correction is exactly unbiased when the effective delay kernel reinjects all of its mass, and carries a bias linear in the unreinjected fraction otherwise; at the no-delay identity kernel it reduces to V-trace. On a tabular Markov decision process (MDP) proof-of-concept, RAC reduces the closed-form policy bias by up to 47.9x at the two-slow-channel configuration, beating wait-for-slow at lower wall-clock cost. RAC integrates with PPO and GRPO through a two-line reward-manager patch.

I describe my solution to the LeHome Challenge 2026, an ICRA 2026 competition on bimanual garment folding. The system placed 1st of 62 teams in the online (simulation) round and 2nd in the real-world final. It improves a vision-language-action (VLA) policy with a reinforcement-learning loop. The policy is its own value function: the same network that predicts actions also predicts success, progress, and a few task-relevant future quantities, and those predictions drive advantage estimation, live failure detection, and candidate selection. The work mostly recombines existing RL ideas with engineering and optimization contributions that can be used together as one recipe or individually: AWR + RECAP combined for flow-matching VLA; an asynchronous distributed training / rollout pipeline through HuggingFace Hub; inference-time hyperparameters optimization via Thompson sampling; a sim-to-real recipe with camera-alignment tooling, heavy augmentation and DAgger-like HIL data collection.

Energy trading decisions depend not only on current market prices, but also on expected future market conditions, and operational constraints. This makes the state representation given to a reinforcement learning agent an important design choice. We study this in HydroDam, a pumped-storage arbitrage environment, using a fixed Double DQN agent. The environment, action space, reward function, network, and training protocol are kept fixed; only the market features are changed. We compare absolute price/calendar features, relative features that compare current prices with recent market history, forecast features, and all combinations of these three feature families. Policies are trained and selected using 2007--2011 Belgian day-ahead prices and evaluated on two test settings: a later same-market test set from 2012--2025 and 39 other ENTSO-E market zones. Absolute features only reaches 28.8% on the test set and a median 5.7% across zones. Relative-only and forecast-only states also stay below a rolling price-score heuristic in the cross-zone median. Combining feature families is much stronger: absolute + relative reaches 49.9% on the test set and a 39.8% cross-zone median, while absolute + relative + forecast reaches 55.6% and 47.5%. These results suggest that state representation is not a minor preprocessing choice in storage-trading RL, but a central part of the policy design: robust transfer requires combining price scale, recent relative price context, and short-horizon forecast information, rather than relying on any single feature family.

Reinforcement learning (RL) has become a prominent framework for developing driving experts in autonomous vehicles. However, most existing RL-based experts are designed to output direct control commands (e.g., throttle, steering), which suffer from a lack of interpretability, high spatial complexity in learning road geometries, and poor compatibility with modern end-to-end planning architectures. To address these limitations, we propose a novel trajectory planning architecture for RL driving experts that integrates an RL policy with a polynomial-based trajectory planner. By employing a Frenet-frame coordinate system, our method simplifies complex road geometries into a curvilinear framework, offering a structured coordinate prior that facilitates policy learning. Furthermore, we incorporate a kinematic feasibility check into the planning stage to ensure that generated trajectories remain within the vehicle's physical limits, effectively mitigating cumulative tracking errors typically found in planning-based systems. We evaluate our approach on key CARLA benchmarks, where it significantly outperforms existing state-of-the-art control-based RL experts. On the CARLA Offline Leaderboard v1 and NoCrash benchmarks, our method improves the driving score by 5% and 11%, respectively, and increases the success rate by 8% and 19%.