Overview

Imitation learning teaches an agent to act by using examples of how a person or expert system behaves, often when specifying a reward is difficult. Behavior cloning directly maps observations to actions, while DAgger adds expert corrections on states the learner encounters, and inverse or adversarial imitation learning infers what behavior to match. Current methods also use diffusion policies, retrieval, world models, and visual, language, or tactile data for tasks such as robot manipulation, autonomous driving, and multi-agent coordination, with benchmarks including LIBERO, RoboMimic, and CARLA. A central challenge is that small errors can lead agents into states missing from demonstrations; improving recovery and safety while learning efficiently from limited, noisy, or conflicting examples remains an open tension.

Latest papers

287 papers37 papers in the last four weeks, up 6% on the four weeks before. 0.3% of all new papers.

Oct 19, 2026

Learn the Task, Not the Schedule: Reducing Spurious Cross-Arm Coupling in Bimanual Imitation Learning

A bimanual demonstration specifies both what the robot should do and one particular schedule for when each arm should act. Standard imitation learning does not distinguish task-required cross-arm constraints from incidental scheduling choices, and may therefore exploit opposite-arm progress as a shortcut cue, creating spurious cross-arm coupling even where relative timing is irrelevant for the current task phase. We introduce Constraint-preserving Temporal Recombination (CTR), a training-time data intervention that preserves recorded arm-local sensorimotor content while retiming the corresponding segments to realize admissible schedules that preserve task-required cross-arm constraints. We use CTR to train Diffusion Policy, Action Chunking with Transformers, and π0.5π_{0.5}, and evaluate them on four simulated and four real-robot tasks. Across six real-robot bimanual progress-state initializations, CTR raises worst-state success from 56% for the strongest non-CTR schedule mixture to 76%. With π0.5π_{0.5} on an Independent task, a 4 s pause transfers 2.56 s of delay to the opposite arm under the Concurrent schedule distribution, but only 0.25 s with CTR. On real robots, CTR violates task constraints in only 0--4% of trials, versus 36--98% for narrow-schedule baselines. Code will be released at https://github.com/MINT-SJTU/CTR.
Oct 19, 2026

NIOCIS: Typed-Decision Next-Hop Routing with Node-Conditioned Option Spaces for Tool-Use Agents

Tool-use agents pay per decision: at each hop a language model re-reads the state and thinks for seconds, though its actions are finite and decisions recur. We present NIOCIS, a routing layer that folds execution logs into a loop-collapsed action graph (forward transitions plus recovery back-edges) answering one typed question: which declared option at a state, plus stop, does the fleet take next? A small model reads a calibrated distribution over them in one pass, emitting no text. On 1,615 production security-operations episodes we measure what the layer replaces, returns and is bounded by. Replaces: the typed encoder matches the logged next hop on 32.2% of the 3,875 held-out decisions (34.9% within the declared options) in 10 ms of encoder time (19.5 ms for the typed-choice form), against 2.2 s per LLM hop; two feature readouts reach 32.8% and 32.3%, an ordering the paired test does not resolve, so the typed layer is chosen for cost, not a separated lead. Tuning: jpt-4b reaches 9.9% zero-shot and 31.9% fine-tuned, a gap an exact paired test does not resolve (p=0.68), at 244 ms. Returns little: the realised hop falls outside the declared options for 7.9% of decisions and outside the model's top twenty for 9.3%; the most confident 15% holds 91.9% accuracy. Bounded: a cross-trace leave-one-out test puts the released state's ceiling at 25.6--44.9% as the state key grows more specific, i.e. repeated states disagree more often than they agree; the option-space ablation attributes the gain to per-node candidate restriction (+3.5 points over a flat catalog) rather than graph traversal (zero within its interval); a gated self-evolution loop lifts every arm above its anchor yet ends below full supervision; a confidence-gated cascade loses agreement when the escalation arm is the zero-shot stand-in; and agreement past a third of the fleet needs a state carrying the plan and the observation.
Oct 16, 2026

DefFlow: Goal-Conditioned Deformable Object Manipulation using 3D Point Flows

Manipulating deformable objects toward desired 3D configurations is a fundamental capability for robots operating in unstructured physical environments. However, as the range of possible initial and target configurations grows, goal-conditioned deformable manipulation requires a robot to infer increasingly diverse manipulation behaviors in addition to precise continuous actions. We introduce DefFlow, a framework for goal-conditioned manipulation of deformable objects toward a broad range of 3D configurations. DefFlow learns from demonstrations by jointly predicting the goal-conditioned 3D point flow of the robot and deformable object. We show that this formulation better captures both the manipulation decisions, such as when and where to grasp, release, or re-grasp, and the precise continuous motions required for goal reaching, rather than predicting only the robot flow. Across rope, granular-media, and cloth manipulation settings with diverse initial and target configurations, DefFlow substantially outperforms goal-conditioned action-prediction and robot-flow-only baselines on success rates and goal-distance metrics. We further demonstrate deployment of our framework on a real robot using real point-cloud observations and unseen goal configurations.
Oct 15, 2026

Diffeomorphic Hidden-State Steering: Improving Imitation Learning Policy Robustness through Expert Hidden-State Geometry

Small observation or action errors can compound during closed-loop execution and disrupt a trained imitation learning policy. We study whether the original expert demonstrations can improve robustness through test-time hidden-state steering without additional demonstrations or policy retraining. Our analysis of a trained policy finds that expert hidden-state variation is concentrated in relatively few dimensions, while the directions capturing this variation differ across regions. Motivated by this geometry, we propose Diffeomorphic Hidden-State Steering (DHS), which learns an invertible nonlinear coordinate map from expert hidden states obtained by replaying the demonstrations through the frozen policy. DHS attenuates coordinates with low expert variance more strongly than those with high variance, then maps the modified coordinates back to hidden-state space. This allows steering directions to vary with the current hidden state while retaining every coordinate. Across four simulated manipulation tasks, DHS improves mean success under observation noise over the unmodified Transformer policies. Controlled comparisons yield higher mean success with learned nonlinear coordinates than with linear coordinates under both hard projection and full-rank rescaling. Combining DHS with action smoothing improves mean success rate under observation noise in certain tasks compared with action smoothing alone.
Oct 15, 2026

X-Fly: Learning to Reach and Stop in Full-Chunk Aerial Vision-Language Navigation

Aerial vision-language navigation (VLN) agents built on large vision-language models are expensive to query. Executing a complete predicted action chunk before replanning amortizes this cost, reducing model inference from 195 to 42 ms per executed waypoint for our policy. It also moves the states at which the policy is queried away from expert demonstrations, which supervise neither recovery from these states nor termination at the learner's own arrival. On OpenFly test-seen, no-stop failures account for 47.78% of full-chunk behavior cloning (BC) outcomes. We present X-Fly, a staged training framework for full-chunk aerial VLN. Recovery DAgger labels learner-visited decision states with smooth chunks that reconnect to future expert segments and with STOP at the first in-region state. Flow-GRPO then jointly refines the chunked flow controller and a Bernoulli STOP policy from same-task closed-loop outcomes, supplemented by outcome-verified stopping supervision. With eight controls per decision, X-Fly reaches 74.18% SR on test-seen and 51.73% on test-unseen, compared with 16.45% and 18.43% for full-chunk BC and 32.04% and 29.46% for DreamFly.
Oct 14, 2026

Learning Navigation from Multimodal Demonstrations with Geometric Candidate Modes

Offline navigation demonstrations are multimodal, because one observation often allows several valid actions, such as passing an obstacle on either side. Behavior cloning averages these actions, potentially leading to unsafe routes. Multimodal policies must discover the alternatives from the data, but similar states can lead to different routes whose current actions overlap. In navigation, the alternatives have a geometric cause and can be computed from one observation. We propose a hierarchical neural-network policy that computes candidate modes from local geometry and learns from demonstrations which mode to follow and how. A high-level policy selects a mode using the goal and its previous choice among a set of candidate modes, and a low-level policy turns it into an action. In simulation, the policy succeeds in over 95% of seen and unseen test cases, compared with at most 43% and 20% for four offline learning baselines and 79% and 67% for a heuristic that uses the same modes without learning. The simulation-trained policy also transfers to a LimX Tron 1 bipedal robot.
Oct 14, 2026

Investigating Behavior Cloning from Few Demonstrations for Autonomous Driving based on Bird's-Eye View in Simulated Cities

This paper investigates the use of Behavior Cloning (BC) for autonomous driving from a bird's-eye view (BEV) perspective in simulated urban environments. BC uses supervised learning to mimic expert driving behaviors. Previous works have applied BC in the CARLA simulator but did not fully address the challenges of traffic light compliance. Our approach enhances BC by integrating a kernel density estimator to adjust training sample weights based on action density, thereby improving the learning of rare but critical actions such as stopping at red lights and accelerating at green lights, specially in scenarios of scarce number of expert demonstrations. Using BEV inputs, which provide an abstract top-down view of the driving environment, our method simplifies the policy learning process. The trained convolutional neural network (CNN) outputs steering and acceleration actions based on these BEV inputs and additional state variables. Experimental results in the CARLA simulator demonstrate that our weighted BC method significantly improves driving performance, achieving higher route completion compared to standard BC. This weighted approach proved to be crucial in learning correct driving behaviors, particularly in test environments not encountered during training, highlighting its potential for enhancing autonomous vehicle navigation.
Oct 13, 2026

Imagining Goals with Geometric Horizon Models for Policy Generalization

Training from offline data has allowed for substantial progress in domains such as robotics, leading to general-purpose policies that can be easily applied zero-shot or efficiently finetuned for downstream tasks. However, these policies can still have poor generalization, both due to the choice of modeling objective and from learning from a static dataset. In this work, we focus on the challenging task of zero-shot goal generalization, where a policy is evaluated on unseen tasks that require combining its existing knowledge (compositional generalization). An avenue for improving a policy's generalization is by generating new experience with world models; however, such generation has proven difficult for longer horizons. Thus, to alleviate this issue, we propose TD-Aug, which samples from a geometric horizon model and allows for directly imagining novel outcomes that can be achieved by composing existing knowledge. We demonstrate that training on these future outcomes as goals for goal-conditioned BC policies significantly improves generalization in stitching-based OGBench tasks.
Oct 13, 2026

What Matters in Quality-Aware VLA Pretraining: A Benchmark and Empirical Study

How should demonstration quality be used during vision-language-action (VLA) pretraining to improve performance after target-task fine-tuning? We compare weighting, filtering, and conditioning across reward definitions and label granularities using 5,000 naturally suboptimal human demonstrations and 24 target tasks. Within each VLA, we match demonstrations and training budgets. Additional comparisons examine pretraining data size and the roles of quality tags during fine-tuning and inference. Across three VLAs, we find that approach-aware episode conditioning improves mean success over plain behavioral cloning; conditioning and filtering favor episode and chunk labels, respectively, on average, with model-dependent reversals; and constant-tag fine-tuning improves mean success after conditioned pretraining with either tagged or untagged inference. We introduce AXIS-Bench to support these controlled comparisons. Built on crowdsourced simulation demonstrations, it provides recorded object and contact states for recomputing quality labels on shared trajectories. Replayable trajectories support workspace-aligned rendering and user-defined real-world target tasks. In our real-world experiment, simulation pretraining raises success from 9.6% to 31.7%. Project page: https://axis-bench.github.io/.
Oct 13, 2026

Where Small Mistakes Matter: Quantifying State Criticality for Imitation Learning

Imitation learning typically treats every demonstration state as equally important, optimizing a behavior-cloning objective uniformly across the dataset. Yet not all states are created equal: in a small number of critical states, even a slight deviation from the expert action can lead to a failure, while in many other states such errors are recoverable and inconsequential. We formalize state criticality in imitation learning as the worst-case degradation in downstream value induced by a bounded perturbation of the expert action at a given state. We then propose SCRIM, a practical State CRIticality Measure that estimates this quantity by training an adversarial disturbance policy and a value critic around a fixed imitation policy. SCRIM assigns each demonstration state a score reflecting how vulnerable the task outcome is to small, policy-relevant action errors. We show that these scores can be used to reweight behavior-cloning losses, improving imitation at consequential states and increasing overall task performance. We further show that SCRIM enables data curation by selecting compact, high-value subsets of demonstrations that preserve performance while substantially reducing the amount of training data. We evaluate SCRIM across various simulated and real-world robotic manipulation tasks, and demonstrate that SCRIM improves policy training, enables efficient data selection, and provides interpretable insights into which moments in a demonstration are most critical for task success. Project website: https://sia-lab-git.github.io/SCRIM/
Oct 12, 2026

Improving Policy Learning from Delayed Interventions

In this work, we explore the idea of expert intervention delays in human-supervised robotic systems for imitation learning system. Previous works in imitation learning have primarily examined the impact of suboptimal experts, where the expert policy makes mistakes, but assumed expert interventions occur immediately after errors are made. Although delayed expert intervention can be viewed as a form of noise that might improve policy robustness, our results show that learning from delayed expert interventions is overall less effective than learning from non-delayed expert interventions; delayed expert interventions are less effective at demonstrating error avoidance behaviors. To address this, we propose Learning from Delayed Intervention (LfDI), which learns a cost function based on where and when the expert intervenes, combined with limited information about the extent of intervention delays provided by the expert. This cost function is then used to infer additional supervision for states without expert data, guiding the policy to avoid states that would likely trigger interventions. We evaluate our method, Learning from Delayed Intervention (LfDI), on simulated environments and show that it outperforms other methods that do not account for intervention delays.
Oct 12, 2026

Imitation Learning of Avoidance Modes for Efficient Multimodal Model Predictive Control

Multimodal model predictive control (MM-MPC) avoids local minima among moving obstacles by solving one optimal control problem (OCP) per combination of avoidance sides, so its computational cost grows exponentially with the number of decision-critical obstacles. We propose SideNet-MPC, which replaces this enumeration with a mode decision learned by imitating MM-MPC as an expert. A graph neural network is trained to predict the expert's minimum-cost mode, one avoidance side per obstacle. Its edges encode, for each obstacle pair, the gap predicted for the moment the vehicle first reaches either obstacle, so a neighbor that blocks a preferred passage can reverse the proposed side. The expert's selection rule compares its minimum-cost mode with the previously executed mode; SideNet-MPC keeps this rule and substitutes the network proposal for the minimum-cost mode. Hence at most two OCPs are solved per control period, and every applied input still comes from an OCP with control barrier function constraints rather than from the network. In 100100 seed-matched closed-loop Monte Carlo runs of mixed traffic, SideNet-MPC reaches 93%93\% of the expert's success rate and needs \SI{18.6}{ms} per control period on a single thread, 48% less than the expert's \SI{35.5}{ms} on sixteen parallel threads. In a controlled gap sweep, its proposed side reverses in the same gap interval as the expert's, whereas proposers without pairwise edges, trained on the same data, never reverse and all deadlock at the blocked gates of a deterministic scene.
Oct 12, 2026

One-Shot Learning for Double-Shot Latte Art: Adapting Cobot Pouring Skills with Dynamic Movement Primitives

Latte art is a complex task requiring dexterity and precision, making it a challenging and commercially relevant task. A one-shot learning system is employed in which a single barista demonstration is segmented into five functionally distinct phases, each encoded as an independent Cartesian Dynamic Movement Primitive (CDMP) with its own parameters. Against a single-CDMP baseline, per-phase encoding reduces transition endpoint error by 3.2-4.5 times and art-stroke shape error by 2.7 times, reproducing jug tilt to below 1 degree RMS in the pouring phases. The learnt pour scales from a 280mL to a 400mL cup through a single spatial offset and a scaled recipe, with no re-demonstration. The phase-based CDMP approach generalises beyond latte art to any demonstration task whose phases have distinct spatial or temporal requirements whilst demonstrating the ability to adapt reproductions to suit different task requirements.
Oct 12, 2026

Imitation Holds a Key to Evolution: Localized Repair and Verified Skill Acquisition for Adaptive Agents

An agent can retain successful trajectories without learning which part of an experience should repair a new discrepancy. We formulate passive imitation (PI) as an architectural loop that localizes a violated contract, extracts a local goal, proposes an effect-matched skill, tests a small patch, and retains evidence after full-task verification. We instantiate this loop as a deterministic symbolic controller and analyze its finite repair-query complexity under explicit identifiability and noninterference assumptions. Across ten simulated tool worlds, PI solves all 600 cold held-out composition tasks with three allowed small trials, compared with 6.17% for a whole-trajectory control, while an equally expressive compositional repair control matches PI. On an unseen tool revision, PI achieves 70.0% success versus 24.0% without small trials and 79.0% with self-trial enumeration; an exploratory 24-task adaptation stream subsequently raises first-submission success to 100%, supporting a limited correction-and-retention mechanism rather than a unique incentive advantage or general intelligence. The CPU-only implementation, integrity checks, and reproducible results are available at https://github.com/FriendOfYishan/PI.
Oct 7, 2026

Contact-Aware Imitation Learning Through Contact Factorization

Generalizable contact-rich manipulation requires robots to preserve intended task behavior while adapting its physical realization to changing contact conditions. However, interaction forces can vary substantially with small changes in surface geometry, orientation, and friction, making policies trained directly on raw force measurements difficult to transfer beyond demonstrated conditions. We introduce FACE, a contact-factorized imitation learning framework that separates intended task behavior from environment-dependent contact factors. Our representation expresses interaction forces in normalized, contact-relative coordinates, while a learned contact-normal estimator and an online friction estimator infer the local contact normal and effective friction scale. Together, these estimators enable force observations to be encoded and policy outputs to be decoded into physical motion and force commands during execution. In this way, FACE adapts execution to current contact conditions while preserving the intended task behavior, without updating the policy parameters. We evaluate FACE on real-robot contact-rich manipulation under unseen variations in surface properties and geometry, demonstrating robust generalization across contact conditions through controlled comparisons with variants that adapt prior approaches to our setting. Videos and additional materials can be found on the project page: https://rcilab.khu.ac.kr/face.
Oct 6, 2026

Teaching a Robot Dog New Tricks: Diverse Quadruped Skills via Combined Reinforcement and Imitation Learning with Adversarial Task Selection

Reinforcement Learning (RL) has enabled legged robots to perform a range of skills in single-task settings. However, applications such as farm robotics or space exploration require diverse skills such as locomotion, digging, or close-range surveying. Training an end-to-end policy to address this problem remains difficult due to challenges such as sample inefficiency and gradient conflict between tasks in multi-task learning. We propose a three-stage method that trains a single policy to perform distinct tasks such as walking, digging, and hopping, and compose them into novel behaviors such as crawling. First, multiple teacher policies are trained using RL on narrowly defined tasks. Then, two additional stages train a student policy with a multi-teacher distillation setup that uses a combined RL and Imitation Learning (IL) objective under an adversarial task selection process that focuses training on the worst-performing task. With this method, we train a student policy that performs 22 tasks using 8 teachers. Evaluations show our method preserves motion quality and tracks commands more accurately than PPO and distill-then-finetune baselines, and in some cases generalizes to new tasks without explicit training. Finally, we demonstrate real-world robustness by deploying the resulting policy on a Unitree B1 quadruped. Video: https://youtu.be/V9yX04EBcFA
Oct 6, 2026

Attacca: Goal-Directed Control under State Continuity for Long-Horizon Embodied Agents

A central capability of embodied agents is to accomplish complex objectives through sequences of interdependent tasks. Yet existing visual goal-conditioned policies underlying these agents are typically evaluated on isolated interactions where the target is already visible, and thus do not capture the conditions that arise during continuous long-horizon task execution. In such settings, each task begins from the state left by the previous one: the agent may end at a different position and orientation, the world may have been modified, and the next interaction target may lie outside the current field of view. As a result, agents relying on such policies may struggle to proceed to the next task when they cannot ground their target in the current observation. To address this challenge, we propose Attacca, a new approach that trains visual goal-conditioned policies on complete search-to-interact trajectories using goal images decoupled from the execution environment. Attacca uses context-decoupled goal sampling to pair each demonstration with a class-compatible masked goal image from another world, removing direct scene and pose correspondence. It learns dense current-view grounding through a target-mask prediction head, providing auxiliary supervision beyond action imitation. We further introduce behavioral-phase conditioning that teaches the policy to distinguish Search, Approach, and Interact stages and adapt its control as execution progresses. We evaluate Attacca on multiple short- and long-horizon embodied tasks in Minecraft. Our method achieves 39.0-47.5% clean success, improving over the strongest baseline by 1.7-2.4x. On long-horizon tasks, it attains 54%, 30%, and 28% completion, yielding up to a 7x improvement.
Oct 5, 2026

Task-Space Imitation Guidance for Efficient Reinforcement Learning

We introduce Task-Space Imitation Guidance for Efficient Reinforcement Learning (TIGER), a reward-construction and pretraining framework for sparse-reward tabletop robotic manipulation. TIGER treats an action-chunked imitation policy not as an executable controller or action prior, but as a local task-space progress estimator: predicted action chunks are converted, using controller-aware action-to-motion mapping, into short-horizon end-effector references, and the RL agent receives dense progress rewards toward these references while the sparse environment reward remains the dominant objective. During pretraining, TIGER uses imitation-guided look-ahead signals to relax conservative value penalties for actions predicted to make task-space progress, reducing off-manifold exploration during early online RL. Across simulation and real-robot experiments, TIGER improves early sample efficiency and reduces measured safety violations while matching or improving final success rates relative to prior RL and IL-RL baselines on the evaluated tasks. Videos and an overview of this work are available at https://sanctuary.ai/news/tiger/.
Oct 5, 2026

Interleaved Projected Gradient Descent for Safe Imitation Learning

We propose an imitation-learning design for neural-network control policies under state and input constraints. Training alternates a standard imitation gradient step with a block of kk safety steps that pull the network's actions toward their projection onto the safe set; at run time, the controller is the trained network alone, with no safety filter. We analyze this scheme as inexact projected gradient descent in the space of policy actions. When the projected actions are recomputed at every safety step and each step moves the actions consistently toward the safe set, letting kk grow logarithmically yields asymptotic constraint satisfaction on the training states and bounds the distance to the constrained optimum of the imitation loss; with the projected actions held fixed, the same holds only if they are exactly representable by the network. On a nonlinear autonomous racing task, we compare our method with adding a weighted constraint-violation penalty to the imitation loss. With a sufficiently large weight, our method matches the lap time of unconstrained imitation while reducing the fraction of violating episodes from 15%15\% to 1%1\%, about six times fewer than the penalty approach at its best weight. Its lap times are less sensitive to the weight, which instead sets how quickly violations vanish during training. In racing, the safety corrections are sparse and the conditions of the analysis do not hold; the gain arises instead through the data collected during training. These gains come at the cost of additional training computation.
Oct 5, 2026

Talk Like You: Imitating How You Speak in Real-Time Talking Head Generation

In daily life, each person exhibits unique speaking habits, leading to subtle yet consistent lip-shape variations even when pronouncing the same word. Although recent talking head generation methods have achieved impressive visual fidelity and lip synchronization, they largely overlook user-specific customization, especially the motion patterns that characterize individual speaking habits. These habits are difficult to model and capture, as their motion patterns are highly fine-grained and often similar across individuals. As a result, many approaches produce overly uniform facial motions and fail to capture diverse, person-specific articulation patterns. To address this, we propose TalkLikeYou, an efficient framework that imitates how a target person speaks in talking head generation. Our method models habit in motion-space and achieves real-time performance through Flow Matching with only one sampling step during inference. We further adopt a two-stage imitation learning strategy to capture subtle distinctions between habits, allowing users to specify a target habit through either a preset style from the dataset or a reference video. In addition, we introduce a new metric PLAD that projects mouth motions onto representative articulation axes to evaluate imitation accuracy and generation diversity. Extensive experiments demonstrate that TalkLikeYou generates high-quality talking heads in real-time and significantly improves speaking habit imitation compared with prior methods. The code is available at: https://github.com/BQ-Wang0511/TalkLikeYou
Oct 5, 2026

Conditional Trajectory Peaks: Single-Pass Multimodal Policies over Action Chunks

Multimodal imitation learning requires diverse executable futures under the same observation and consistent behavior across replanning cycles. We present Conditional Trajectory Peaks (CTP), a single-pass policy framework that jointly predicts complete action-chunk candidates, probability masses, and trajectory scales. Distribution-Aware Peak Specialization (DAPS) specializes trajectory peaks using trajectory-level posterior responsibilities and mass- and scale-modulated overlap constraints. Evidence-Gated Trajectory Belief Transport (ETBT) maintains cross-chunk consistency through geometric correspondence between exchangeable candidate sets, while allowing current policy evidence to override historical constraints. CTP achieves a coverage score of 91.40% on Push-T; success rates of 100.0%, 79.72%, and 84.44% on D3IL Avoiding, Aligning, and Sorting-2, respectively. On LIBERO, CTP achieves an average success rate of 97.25%. In real-world dual-arm experiments, CTP preserves both placement modes in a two-plate task, succeeding in all 50 trials. On bottle uprighting and pen placement into a holder, it maintains success rates comparable to π0.5π_{0.5} while reducing policy inference latency from 218.24 ms to 75.80 ms. These results demonstrate that single-pass trajectory modeling can combine multimodal behavior, closed-loop consistency, and efficient inference.
Oct 5, 2026

Behavioral Cloning Mystery

Behavioral cloning (BC), despite its simplicity, exhibits many counterintuitive phenomena in the real world. For example, the performance of BC often keeps increasing as the model overfits more to the dataset, and fully closed-loop policies often completely fail without action chunking. Unfortunately, properly studying these anecdotal phenomena ("behavioral cloning mysteries") is challenging: in the real world, datasets and experiments are costly and not fully controllable; in simulation with synthetic data, these phenomena are often not easily observed partly due to the discrepancy between scripted policies and human demonstrations. In this work, we propose OCBench, a robotic manipulation benchmark with controllable scripted policies that have similar properties to human demonstrations. We show that, by mimicking key properties of human demonstrations, OCBench reproduces many anecdotal BC-related phenomena in controlled settings. With its GPU-accelerated environments and scripted policies, we demonstrate how OCBench enables scientific studies of previously reported BC-related phenomena by analyzing and refuting various hypotheses. Project page: https://seohong.me/projects/ocbench
Oct 5, 2026

Bilinear Flow Policy: Distributional Extrapolation for Goal-Conditioned Visuomotor Imitation

Goal-conditioned imitation learning (GCIL) with flow matching is a promising framework that can represent multimodal behaviors while adapting to diverse, user-specified goals, yet often fails when goals lie outside the demonstration support. To extrapolate to such unseen goals without collapsing multimodality - a problem we call distributional extrapolation - we introduce Bilinear Flow Policy (BFP), a generative visuomotor policy that combines transductive retrieval with a bilinear conditional flow. Given an unseen observation-goal pair, BFP retrieves an "anchor" training example and transductively reformulates the unseen pair as this familiar anchor plus a residual term. For this decomposition to guide action prediction, the residual must compactly encode how the current observation-goal pair differs from the anchor, and the anchor must be chosen so that this difference is predictive of the corresponding action distribution. BFP achieves this with pretrained visual features and a novel learned anchor-selection algorithm. The novel bilinear flow then models how the anchor and the residual jointly determine the multimodal action distribution. We prove that, for bilinear flow under suitable assumptions, action distribution error at unseen goals is bounded by the in-distribution flow-matching error up to problem-dependent factors. Across five manipulation tasks in simulation, BFP achieves 2.63x the out-of distribution success rate of a GCIL policy and 1.36x that of the strongest extrapolation-targeted baseline. On two real-world tasks, BFP improves over GCIL by 32%. Finally, our theory yields practical, pre deployment diagnostics for predicting which trained policies will extrapolate well and to which unseen goal.
Oct 3, 2026

DreamFormer: Dream Imitation with a Transformer World Model for Language-Conditioned Robotic Manipulation

We introduce DreamFormer, a model-based agent that acquires language-conditioned, multi-task skills by imitating expert demonstrations within the latent imagination of a learned world model. DreamFormer first learns a task-agnostic Transformer world model from unstructured play data, then acquires task-specific behaviors by optimizing an intrinsic reward that aligns agent-generated rollouts with expert demonstrations in latent space. Since the policy is trained on-policy inside imagination, it is exposed to its own errors during training, mitigating the covariate shift inherent to offline behavioral cloning. To make long-horizon imagination affordable, DreamFormer encodes a high-resolution multi-view robotic observation into a single input token, avoiding both spatial downsampling and the multi-token representations used by prior Transformer world models. On the long-horizon CALVIN benchmark, DreamFormer outperforms LUMOS, the comparable model-based agent, on single-environment evaluation (2.52 vs 2.34 average tasks completed per chain of five) while keeping imagination rollouts tractable. Against HULC, the behavior cloning baseline, it nearly doubles performance on zero-shot transfer to an unseen environment (1.30 vs 0.67), indicating that dynamics learned by the world model transfer more readily than a directly cloned policy. This is consistent with the role attributed to internal models in biological agents, where a model of environment dynamics supports behavior in situations not previously encountered.
Oct 1, 2026

Faynt: Scaling and Optimizing Policies for Competitive Melee

We introduce Faynt, a family of 10M- and 75M-parameter Transformer policies for Super Smash Bros. Melee, each controlling all 26 characters with a single checkpoint. After reinforcement learning (RL), the 10M wins 240 of 244 same-character games (98.4%) against fourteen specialist and multi-character releases on their supported rosters, with a winning record against every release. These opponents retain 21- or 24-frame action delays; Faynt uses no added delay, and we have not isolated the effect of this difference. In a separate evaluation against a privately supplied zero-delay Slippi-AI model, the 10M wins all 68 games across two conditioning settings. We study architecture, optimization, scaling, and hyperparameter transfer to guide pretraining on approximately 840,000 human replays. Post-training combines rank- and outcome-based curricula, 75M-to-10M distillation, and RL restricted to Fox mirror matches. On the initial 152-game benchmark, the supervised 10M wins 69.7% of games, compared with 45.4% for the pretrained 75M, despite higher overall held-out controller-prediction loss. The weighted validation loss used for supervised checkpoint selection agrees with the win-rate ordering of all four pretrained and supervised policies. After supervised post-training, both models take less damage per minute, build larger early leads, and win more often after losing the first life. Optimized inference on recorded game states averages 5.2 ms per decision for the 10M and 8.7 ms for the 75M on an NVIDIA T4, excluding emulator execution and communication. We open-source the weights, both benchmark suites, and a platform for automated model tournaments.
Oct 1, 2026

Iterative Policy Refinement through Semantic Rollout Analysis

Structured policies improve efficiency, robustness, and interpretability in imitation learning by introducing task-specific inductive bias, but existing structure generation methods rely either on extensive human input or on static domain knowledge encoded in LLMs, which may be inconsistent with the expert demonstrations. We propose a closed-loop framework that iteratively refines structured policies using LLM-guided analysis of policy rollouts. By logging rollouts as semantically meaningful tabular data and prompting the LLM to generate diagnostic analysis code, our method identifies suboptimalities in the policy structure and iteratively corrects them without requiring human instruction. Experiments on car racing and door opening tasks show that our approach improves imitation learning performance by up to 15% over zero-shot LLM-generated structures and requires 75% less compute to achieve the same reinforcement learning performance. These results demonstrate that tabular rollout analysis provides an effective feedback signal to align LLM-generated policy structures with expert demonstrations, and we can utilize it to generate good policy structures automatically.
Oct 1, 2026

Measuring the Stability Assumption Behind Action Chunking

Action chunking improves the performance of policies learned by behavioural cloning, and several mechanisms have been proposed to explain why, including temporal consistency, horizon reduction, representation learning, and reduced error compounding. We instead study what happens to an action error once it enters the system. At each state, we inject a small action error and measure how fast it grows or shrinks under two execution regimes: open-loop, where the rest of the chunk is replayed without replanning, and closed-loop, where the policy replans after the perturbation. The fitted rate labels each state as contracting, expanding, or unresolved. Across twelve manipulation tasks from three benchmark suites, we find that confidently stable states are rare, while error amplification is common among states whose propagation rate can be resolved. We further find that the measured propagation rate depends strongly on the fitting horizon: amplification is typically front-loaded, so short windows can overestimate longer-horizon propagation. Finally, we train predictors on these labels and find that a state's open-loop regime can be recovered from camera frames and proprioception alone, while its closed-loop propagation is only partially recoverable because it also depends on how the policy acts after the perturbation. These results suggest that error-compounding arguments alone do not provide a complete account of action chunking: neither passive open-loop dynamics nor policy replanning consistently contracts an injected error, and replanning rarely turns open-loop amplification into confident contraction. This suggests that closed-loop reactivity should be trained explicitly, using perturbation- and tree-coverage-oriented training to expose policies to deviations they must recover from, rather than expected to emerge reliably from standard imitation learning.
Oct 1, 2026

Experience-Based Feasibility-Aware Generative Adversarial Imitation from Observation under Embodiment Mismatch

With the increasing use of robot-free demonstration interfaces that provide state trajectories without action labels, imitation from observation has become a promising approach for learning robot behaviors from human demonstrations. However, due to differences in embodiment and dynamics between humans and robots, demonstrated human motions may not be feasible for the robot, potentially degrading policy performance. In this study, we propose Experience-Based Feasibility-Aware Generative Adversarial Imitation from Observation (EF-GAIfO), which estimates the feasibility of state-only demonstrations from the robot's own experience rather than relying on explicit dynamics models or large prior exploration datasets. A key feature of EF-GAIfO is that the notion of feasibility evolves with policy learning: as the policy improves and the robot experiences a broader range of state transitions, the feasible region is progressively expanded, allowing additional demonstrations to be incorporated into learning. This enables feasibility-aware imitation that adapts to the current stage of policy learning, rather than relying on a pre-designed feasibility criterion. We validate the effectiveness of EF-GAIfO on a locomotion task in simulation and on a real quadruped robot performing a object-reaching-and-grasping task.
Sep 30, 2026

RoboCoach: World Models as Active Coaches for Compositional Robot Skills

Long-horizon robot manipulation reuses skills across many task compositions, but improving these compositions with additional end-to-end demonstrations is costly. A practical self-improving system must decide both what to teach next and where to apply that supervision. We present ROBOCOACH, a world-model-guided coaching framework that uses imagined failures to guide demonstration requests and expert updates. Its Route-Imagine-Diagnose-Improve (RIDI) loop executes reusable skill experts inside COACHWORLD, our shared action-conditioned world model, and uses a progress judge to record the first subtask that fails to complete. Aggregated records select which subtask demonstrations to acquire and which expert adapters to update. Across two simulation suites and two real-robot platforms, imagined and deployed success correlate over 22 task-policy pairs (rho = 0.840). Controlled comparisons show that our coaching method outperforms matched baselines under matched data budgets and update schedules. With only 150 additional subtask demonstrations, success rises from 13.3% to 75.0% on Franka and from 40.0% to 83.8% on AgileX. The coached experts also transfer to four held-out compositions, achieving an average success of 35.0%, compared with 0% for a shared-policy baseline updated with uniformly acquired demonstrations. Together, these results show that world models can serve as active coaches, turning imagined failures into targeted supervision for modular policy improvement. Project Page: https://robocoach-ai.github.io/
Sep 30, 2026

Plan-Conditioned Imitation for Robust Object Retrieval under Self-Occlusion in Dense Clutter

Retrieving objects from dense clutter requires rearrangement during which the manipulator can occlude objects while moving them. Repeated arm withdrawals to restore visibility interrupt execution. We introduce TRACE, a plan-conditioned imitation framework for retrieval under self-occlusion. A single unoccluded observation initializes a digital twin, where a privileged teacher generates a fixed nominal rollout. A recurrent student combines local rollout context, partial object observations, and proprioception to select actions that can correct deviations from the prediction. Behavior cloning initializes the student; DAgger refines it with teacher labels on student-visited states. The rollout remains fixed throughout execution, so the deployed student needs neither online teacher queries nor additional simulator rollouts during pushing. On 511 simulation test scenes, TRACE achieves 90.7% success versus 43.4% for nominal replay and 96.7% for the privileged closed-loop teacher. At a matched 26,373-label budget, student-state supervision achieves 87.8% versus 66.7% for expert-only cloning, demonstrating gains beyond additional labels. On a UR5e, TRACE achieves 90.0% success versus 95.0% for the closed-loop teacher, while reducing total execution time from 192.7 s to 67.3 s. It avoids the teacher's 16.8 sensing-related arm retractions per trial during pushing, retaining a final withdrawal for graspability evaluation. Code and data will be released at: https://trace-retrieval.github.io.