Long-Horizon Robotic Manipulation

Real-time robot control demands enough visual history to infer motion and task progress, but processing that history can delay action. We present Long-WAM, a model-system framework for scaling the context of causal world-action models under real-time control constraints. Our central finding is that access to history is not the same as using it: longer histories pay off far more when the video foundation is pretrained autoregressively (AR). We first learn causal prediction from robot and egocentric videos without action labels, then preserve this history-to-future structure during world-action adaptation. On RoboCasa GR-1, increasing context from 0.0 to 19.2 seconds raises success from 63.3% to 78.7%, whereas a bidirectionally pretrained initialization shows no net gain; robot-domain AR pretraining further raises peak success on GR-1 and LIBERO-Long. Long-WAM also achieves the best results among compared methods on LIBERO-Long, RoboTwin 2.0, and DOMINO. Streaming observation encoding, asynchronous execution, and hardware-specific acceleration enable deployment on RTX 5090, DGX Spark, and Jetson AGX Thor without dropping future prediction; on RTX 5090, each action chunk, including future-video latent prediction, takes 107.4 ms. Real-time deployment on Unitree G1 and YAM supports dynamic and long-horizon manipulation, including 95% success on dynamic cup stacking, where Pi0.5 and Fast-WAM succeed in none of 20 trials. As a memory-informed executor, Long-WAM also complements higher-level planning in composite tasks.

We present FoldBack, a self-correcting masked generative policy for long-horizon garment folding. Existing long-trajectory policies may continue after a missed or slipped grasp even when the garment has not reached the intended configuration. We structure FoldBack's recovery mechanisms around three inference-time decisions: when to refine and verify, how to roll back, and where and how to retry. FoldBack aligns refinement and grasp verification with pick-and-place events, returns the robot to a retryable pre-grasp configuration while preserving successful grasps, and selectively regenerates the failed segment and selected future actions while avoiding previous failed grasp locations. To our knowledge, FoldBack is the first editable full-trajectory policy to unify these decisions, enabling failed interactions to be detected, undone, and repaired before execution continues, without recovery demonstrations or base-policy retraining. Across 33 real garments from six categories, FoldBack achieves 75.2% final folding success and 0.837 final-mask IoU, versus 45.7% and 0.689 for the strongest prior baseline.

Achieving human-level dexterity in complex, unstructured environments requires the seamless integration of whole-body scene interaction and dexterous object manipulation skills. While existing physics-based controllers generate physically plausible behaviors in each domain, they largely address these two capabilities independently. In this paper, we present Co${}^{2}$Skill that integrates scene interaction and dexterous manipulation through a unified policy formulation. Built on a pretrained motion prior, the policy uses task and phase dependent observation masks to select information relevant to the current interaction goals. We introduce a goal-conditioned loco-manipulation curriculum that combines partial reference guidance for precision with exploration from varied initial states while allowing goal-directed execution beyond the demonstrated trajectories. We further introduce a cross-task curriculum that jointly trains individual skills and selected task sequences, preserving physical states across task boundaries and maintaining grasps during subsequent scene interactions. Together, these support sequential task execution and simultaneous scene interaction with object manipulation. We evaluate sitting, standing, climbing, stair traversal, and goal-directed manipulation, together with sequential execution and with random different conditions. Additionally, we demonstrate skill compositions in indoor environments, illustrating their integration within the same control formulation.

Cluttered indoor environments, where large and heavy objects are scattered across diverse surfaces, require humanoid robots to sequentially navigate, grasp, transport, and accurately place each item at its target location within a single uninterrupted episode. This long-horizon, whole-body loco-manipulation task remains a significant challenge for current methods. Previous approaches often suffer from two main issues: easy-reward bias, where training overemphasizes early transport stages at the expense of later ones, and catastrophic forgetting, where focusing on later stages leads to a decline in earlier-stage performance. In this work, we introduce Humanoid Horizon, a unified policy framework designed to overcome these limitations through three interrelated mechanisms. The Parallel Training Strategy organizes $N$ scenes into $S$ concurrent stage streams governed by a shared policy, ensuring all transport stages receive continuous gradient updates and removing the bottleneck of sequential optimization. The Dynamic Starting Mechanism updates each environment's initial state with terminal states from upstream rollouts, gradually broadening transition coverage and enhancing robustness at stage boundaries. Reward Gating sets the reward to zero for the rest of the episode in later-stage streams when the immediately preceding object is displaced beyond a set threshold, so the shared policy learns not to disturb a just-placed object and earlier placements are preserved throughout the episode. Collectively, these strategies achieve per-stage success rates exceeding 80% on the two-object LHM-Humanoid benchmark (350 training scenes, 66 held-out scenes). As the number of sequentially transported objects grows beyond two, success declines with the horizon, but the degradation is graceful relative to the sharp drop seen in all baselines.

Demonstration-conditioned policies provide a natural interface for specifying robot behavior, yet long-horizon manipulation remains difficult when visually similar states recur across different stages or when demonstrations and executions proceed at different speeds. We identify the resulting failure mode as stage confusion and introduce Progress-Aligned Context for Execution (PACE), a stateful method that continually reinterprets a complete demonstration according to realized execution progress. PACE compresses the demonstration into ordered multimodal prompt tokens and uses training-only dual-edge attention supervision to expose its latent stage structure. During execution, an episode-local fast-weight memory causally encodes realized action-observation transitions and modulates prompt cross-attention, producing a progress-aligned context for a unified diffusion action expert without test-time stage labels or stage-specific policies. PACE improves success from 88.9% to 94.0% on LIBERO-Gen Goal Chain, from 79.1% to 83.3% on Spatial Combination, and from 33.3% to 73.3% on the two-step Block Routing tasks. Failure analysis further indicates that structured demonstration alignment and causal execution memory jointly mitigate stage confusion.

Most vision-language-action (VLA) models rely on current observations and lose task-relevant evidence once it leaves view, limiting performance on long-horizon, memory-dependent tasks. Existing efforts incorporate compressed historical features or sparse visual keyframes. However, isolated snapshots can leave the policy uncertain about what changed during past interactions and which action should follow. To overcome this limitation, we propose EvoMem-VLA, which constructs state-evolution memory by explicitly encoding and retaining observed changes between historical states. These change representations preserve evidence of interaction outcomes, allowing the policy to track task progress beyond isolated snapshots. Specifically, we introduce conditional delta tokenization to encode ordered frame pairs into directional, source-conditioned delta tokens, each associated with its corresponding state evidence. A shared VLM backbone supports task-adaptive routing: normal long-horizon tasks follow a direct action route, whereas multi-stage tasks use a subtask route that generates an executable subtask as an additional input for action generation. With a single jointly trained policy for each simulation benchmark, EvoMem-VLA achieves success rates of 80.7% on RMBench, 82.0% on RoboMME and 83.8% across four real-world tasks spanning two robot embodiments. These results represent substantial improvements over the previous state of the art in all three evaluation settings.

Although the advancement of vision-language models (VLMs) has endowed robots with enhanced environmental understanding and task reasoning, a comprehensive evaluation methodology is important to advance the integration of VLMs in robotic navigation and manipulation. However, current benchmarks lack a comprehensive method to evaluate diverse robotic tasks, and evaluation metrics remain relatively constrained, making it difficult to assess the embodied capabilities of VLMs in a thorough and fine-grained manner. To address this issue, we propose RMMBench, an evaluation benchmark that requires robots to understand language instructions and perform long-horizon tasks in continuous spaces. RMMBench seamlessly integrates high- and low-level embodied tasks into a unified framework, constructing a "navigation-manipulation" task suite comprising 70 canonical task scenarios that range from localized manipulation to long-horizon composite navigation. The results reveal that leading VLMs still face major challenges in spatial localization when performing mobile manipulation tasks, and also highlight the necessity of enhancing the spatial perception capability of robots during long-horizon interactions. RMMBench can be accessed at https://mxxq-stack.github.io/rmmbench-project/

We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining. Our key insight is that a broad controller already holds much of the competence a new task needs, and that this competence becomes accessible through an interface between planning and control that is expressive enough to specify contact-rich, multi-stage interactions, yet executable and measurable enough that execution feedback can guide planning from experience. InterEvolve realizes this interface with two components. First, we develop an object-aware forward-backward (FB) behavioral foundation model, whose object residuals on a frozen body prior turn a new reward about the body or objects into loco-manipulation behavior at test time. Second, we specify tasks as reward programs: staged rewards with completion conditions and tunable constants. A large language model (LLM) agent revises the program structure in context, drawing on execution feedback and a skill library of verified programs, while a numerical optimizer tunes its constants. With every candidate verified across parallel simulation scenarios, the program explores new ways to induce, repurpose, and compose the controller's existing motor competence for the task at hand, and thus improves over iterations. Experiments show that human-designed rewards leave much of the FB model's loco-manipulation competence untapped, whereas the programs InterEvolve evolves release it, sometimes through novel strategies. It further produces behaviors for diverse tasks, complex scenes, and long-horizon compositions in simulation, and evolved skills run autonomously on a physical Unitree G1 from egocentric onboard perception.

Vision-language-action (VLA) models struggle on history-dependent manipulation tasks, where the current observation alone does not determine the action, and the policy needs a memory of the history. Existing memory methods decide what to remember by design, for example, keeping frames with large pixel changes, and show inconsistent gains across tasks. We view what to remember as an optimisation problem. From the POMDP formulation of imitation learning, we show that the optimal memory maximises the conditional mutual information $I(a_t; m_t \mid o_t)$ between the action and the memory given the current observation. Intuitively, this means preserving the action-relevant information in the history that is not already contained in the current observation. Based on our analysis, we propose Divide-and-Remember (D&R), a recursive memory method that learns a memory function $m_t = M(h_t)$ and scales to long contexts while staying compute-light. It involves two strategies: (1) the selection over the full history is divided recursively into subproblems of top-$K$ selection over $2K$ tokens, so that fixed-size, lightweight selectors learned end-to-end support an unbounded history; (2) all recursion blocks share one selector, which captures the selection rule common to every block and keeps the method efficient. On RoboMME, a benchmark of 16 long-horizon manipulation tasks that require remembering when, where, what, and how to act, D&R achieves a state-of-the-art average success rate with consistent gains across all four suites under a budget of only 64 tokens; real-robot experiments show the same gain. Code, checkpoints and more results are at https://dnr-memory.github.io/

A robot may lose sight of an object it must later retrieve, need to recall what a person demonstrated earlier, or track which steps of a task it has already completed. Current vision-language-action (VLA) policies often fail once the information needed for action disappears from the current observation, making memory critical for long-horizon robot behavior. Existing approaches typically provide longer histories or learn implicit memory from observation-action trajectories. But action supervision tells a policy how to act, not what to remember: it does not specify which past facts should persist or how they should change as new evidence arrives. We therefore separate maintaining an evidence-grounded account of the past from learning how to act on it. This insight motivates Explicit Concept Memory (ECoMEM), which represents task-relevant history with a reusable library of grounded concepts. An evidence-based Writer selects and updates these records, while a learned Reader turns them into memory tokens that directly condition the VLA. Across 16 RoboMME tasks, ECoMEM leads the evaluated robot policies on 15 tasks. On two new real-robot tasks, the same memory library either transfers directly or requires only one new concept, achieving 86.1% success versus 8.6% for a no-memory VLA. These results show that explicit concepts provide a reusable and extensible memory interface for robot control. Project website: https://ecomem.github.io/

Vision-language-action policies often see only one or a few recent frames, which makes it difficult to evaluate how they use information that disappears during a task. We introduce MIKASA-Robo-VLA, a benchmark of 90 language-conditioned manipulation tasks. All but 10 hide the cue an action depends on. Those 10 are reactive controls. MIKASA-Robo, the suite it rebuilds, has 32 tasks and uses language only in a representative VLA subset. Here every task provides an instruction, while memory-dependent tasks hide a task-relevant cue and reactive controls keep it available. For 70 tasks, environment phase timings specify an information gap, and for 28 of them the gap exceeds the 16-frame window of the widest fixed-context VLA we survey. The gap counts only the interval the cue is provably absent, not the full duration a policy must retain it, so every memory-dependent task still requires memory by construction, including the ones whose measured gap is short. We release 22,500 oracle trajectories across 10 memory types in RLDS and LeRobotDataset v3. A reference $π_{0.5}$ baseline with current images and proprioception, but no observation history or explicit memory module, is fine-tuned on 14 tasks and achieves 0.211 $\pm$ 0.044 mean task success. Its lower success on the evaluated Long-split tasks is confounded by open-loop chunking and the memory types represented in that subset. Project page: https://mikasarobo.github.io/

Pretrained robot policies provide useful action priors, but long-horizon manipulation still requires coordination between semantic reasoning and physical execution. Semantic reasoning operates at a coarser timescale than physical interaction, while episode-level failures provide limited guidance on which system component should be revised. We propose DynaHarness, a dynamic physical harness that couples semantic reasoning with physical governance through a shared execution contract and turns failure evidence into validated capability revisions. To be more specific, the slow brain proposes capabilities and symbolic arguments, while the fast brain grounds and monitors commands, refuses unresolved actions, substitutes capabilities, and requests replans when needed. The physical execution contract bounds each accepted command and records execution evidence across analytic skills, recovery skills, and the frozen VLA. Failure attribution localizes faults in these records and directs targeted revisions of reusable capabilities or execution mechanisms. Paired regression checks govern admission or rejection, closing the self-evolution loop. On LIBERO-Pro, DynaHarness achieves 75.2% on 800 newly sampled initial states, compared with 17.5% for the frozen policy. With the same capability library, full dynamic execution reaches 74.0% versus 63.9% under nominal one-step replanning. This demonstrates the value of DynaHarness as a dynamic physical harness that governs how existing capabilities are grounded, monitored, and coordinated during execution. Our project page is at https://denghaoyuan123.github.io/Dynaharness_page/.

Robot foundation policies predict action chunks, but how many actions to execute before replanning depends on the current task phase. We introduce ChunkTrust, which treats the execution horizon as a latent variable inferred from action-expert evidence rather than a fixed hyperparameter. Its training-free Action-aware Horizon Selector (AHS) combines intra-chunk spectral stability of generation traces with inter-chunk continuity between executed history and predicted actions. An online Beta posterior with kernel forgetting tracks horizon preferences across replans. A lightweight Query-based Horizon Adapter (QHA) optionally learns a context-conditioned dense prior from complementary evidence, fused with current evidence and episode-local Beta memory while the base policy remains frozen. Across RoboTwin2.0 and RoboCasa GR1 Tabletop, AHS improves overall task-averaged success for each evaluated base-policy configuration, including gains of +6.80 percentage points on $π_{0.5}$ over all 50 RoboTwin2.0 tasks and +9.67 percentage points on Qwen3GR00T in RoboCasa. AHS+QHA raises the gain over Base to +9.44 percentage points on the eight-task $π_{0.5}$ evaluation. On four real-world household tasks, AHS improves the equal-task mean normalized process score from 50.4% to 57.5%. Ablations examine the contributions of both evidence terms, temporal memory, and the learned prior. Project page is https://hf618.github.io/ChunkTrust.github.io/

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/

General-purpose agents can plan, use tools, and revise their behavior from feedback, but it remains unclear whether these capabilities transfer from digital environments to embodied manipulation. To investigate this question, we introduce LIBERO-Agent, an agent-native benchmark for evaluating these agents in robot manipulation tasks. Rather than asking agents to submit task-level Python control programs or operate through high-level robot skills, LIBERO-Agent provides an interactive robotic environment where agents can select which observations to inspect, process them with their own tools, and issue native action commands. LIBERO-Agent integrates 200 tasks into a common interaction framework and provides a 30-task primary suite that separates perception, short-horizon execution, and long-horizon composition. Results reveal a pronounced reliability gap: while agents perform well on perception and easy short-horizon tasks, their performance degrades substantially on hard short-horizon and long-horizon tasks. Richer observations improve short-horizon manipulation, while demonstration benefits depend on the agent and format. Among these agents, GPT-6 Astra achieves the strongest overall performance. Further analysis shows its major advantage lies in mechanism interaction, especially when sustained physical contact is needed, while its remaining failures stem from cross-stage interference and geometric errors.

Egocentric human demonstrations offer an accessible source of task experience, but differences in body scale and controller response, together with missing robot states, limit their value as humanoid training supervision. We present EgoAlign, a data-construction framework that converts these demonstrations into action and state supervision compatible with a general-purpose, continuous whole-body controller, without collecting physical-robot demonstrations. Using the target-robot model and simulator, EgoAlign guides demonstration collection through execution feedback. It preserves locomotion references for visually guided periodic stepping while adapting upper-body interaction geometry through scale alignment and controller-in-the-loop refinement. A final causal replay reconstructs the corresponding robot states and motion-token labels for training with the human observations. We assess the resulting supervision by fine-tuning a vision--language--action model solely on adapted human demonstrations and deploying it zero-shot on a physical humanoid. The resulting policies perform long-range object relocation, navigation to unseen goal positions, and independently evaluated foot interaction. Refinement improves simulated hand alignment and physical pickup success over kinematic alignment alone, while human collection reduces on-site acquisition time relative to teleoperation. https://lambdahumanoid.github.io/EgoAlign/

Long-horizon robotic tasks are vulnerable to unexpected environmental changes that can render planned actions ineffective or unsafe. To address this, robots must detect such changes as they occur, interpret their impact, and adjust their actions accordingly. Traditional rule-based decision-making pipelines are brittle in open-world conditions, as they are hand-tuned for specific scenarios and lack generalization. Vision-Language Models (VLMs) offer a promising alternative as they combine broad world knowledge with unified visual--text reasoning, enabling them to generalize across diverse scenarios and generate accurate, grounded task plans. However, for effective deployment in dynamic real-world settings, VLMs must be embedded into frameworks capable of handling uncertainty and environmental changes. Existing frameworks broadly address this reactively, triggering replanning only after execution failures or post-task checks, risking failed actions. Some methods verify conditions before actions, but these discrete checks miss changes occurring during execution. To address this, we present ProAct-VLM, an adaptive, physically grounded task planning framework that integrates VLMs within a real-time perception--feedback loop. ProAct-VLM continuously monitors the environment and re-plans as soon as relevant changes are detected, enabling adaptation before failure occurs. Evaluations against multiple baselines and across different VLM backbones show that our framework improves both success rates and efficiency in dynamic, long-horizon manipulation tasks. Project page: https://github.com/moured/ProAct-VLM

Vision-language models can coordinate long-horizon robot manipulation, yet successful task reasoning still depends on whether local physical interactions produce the intended effects. We study how repeated interaction can improve this capability without updating the base model. We introduce RoboHarn-Evo, a dual-loop harness that evolves Hierarchical Physical Knowledge (HPK) from physical experience. HPK couples two levels of reusable knowledge: Task Knowledge captures which subtask should be executed and when it is complete, while Action Knowledge captures object-relative geometric strategies and their physical effects. During execution, the agent retrieves knowledge at the corresponding decision level and grounds it in the current scene under the task goal. Across episodes, physical feedback is used to revise historical knowledge, update its applicability, and organize reusable entries for subsequent retrieval. Experiments on RMBench show that HPK improves average success by up to 24.2 percentage points across different agent models. With 80 interaction rollouts, held-out success rises from 48.3% to 75.0% for GPT-5.5 and from 70.0% to 88.3% for GPT-6. RoboHarn-Evo also resolves over 83% of historical knowledge errors while retaining 95.8% of valid knowledge, and transfers zero-shot from RMBench to RoboDojo with gains of 35.0 and 25.0 percentage points. These results demonstrate that physical interaction can be accumulated into reusable knowledge for improving subsequent manipulation.

Learning robust manipulation policies for diverse, long-horizon tasks from limited demonstrations remains a fundamental challenge in robotics. We present DROM, a language-guided diffusion framework that enables robots to learn, represent, and compose multiple manipulation skills within a single generative policy. DROM leverages Dynamic Movement Primitives (DMPs) to augment a small set of expert demonstrations into expressive multi-skill datasets, substantially reducing data collection while improving spatial generalization beyond the demonstrated workspace. Building upon Motion Planning Diffusion (MPD), we extend the diffusion architecture to support language-conditioned multi-skill trajectory generation through cross-attention, allowing a single model to generate skill-consistent motions for a diverse set of manipulation primitives, including orientation-sensitive behaviors that are difficult to design using conventional motion planning or hard-coded controllers. For long-horizon manipulation, a large language model decomposes high-level operator requests into executable sequences of skills, enabling natural language interaction and autonomous task execution. We validate DROM on a Franka Emika Panda robot, a FANUC CRX25ia robot, and in MuJoCo simulation across a wide range of manipulation tasks. Experimental results demonstrate that DROM outperforms Motion Planning Diffusion and Behavior Cloning baselines, achieves robust multi-skill generalization, and composes learned skills to reliably execute long-horizon manipulation tasks from natural language instructions using only a limited number of human demonstrations. Datasets, simulation environments, and more at https://github.com/automation-robotics-machines/drom.

Vision-language-action (VLA) models have driven rapid progress in robotic manipulation, demonstrating strong fine-grained control and promising performance on long-horizon tasks. However, many existing VLAs lack explicit access to interaction history, making them vulnerable to perceptual aliasing: similar current observations and robot states at different task stages may induce action ambiguity and lower success rate. Existing methods incorporate temporal or progress cues through feature conditioning, action-prior modification, or sampling guidance. However, methods that jointly fine-tune memory modules and the base VLA incur additional policy-training costs, motivating the separation of trainable history-conditioned steering from frozen base-policy refinement. We propose ActMem-VLA, a dual-expert handover architecture that augments a frozen, fine-tuned VLA with a memory plugin comprising a Mamba-based memory module and a lightweight PreAction Expert (PAE). Specifically, Mamba encodes executed-action history into memory that conditions PAE alongside current context. With these inputs, PAE steers task progression during early, high-noise denoising, then passes the partially denoised action to the frozen Action Expert (AE) to refine action details during the remaining low-noise steps. The fine-tuned base VLA remains frozen throughout training, while only the Mamba module and PAE are jointly optimized. On LIBERO-Mem, ActMem-VLA achieves 80.8% average success across all ten tasks, compared with 65.2% for $π_{0.5}$ and 49.5% for MemoryVLA, while introducing only 3.45% additional parameters. Across four real-world tasks, it improves the average success rate over $π_{0.5}$ by 28.8%.

World-Action Models (WAMs) improve robotic manipulation by conditioning action generation on predicted future observations, but future prediction adds further inference overhead to already expensive iterative action generation. Action chunking can amortize this cost over multiple actions, yet performance degrades over long execution horizons because later actions remain conditioned on stale observations. We introduce STAIRCASE POLICY, a streaming inference and training framework that turns a flow-matching VLA into a JEPA-style WAM and partitions a large action chunk into sub-chunks at staggered denoising stages. Near-term actions are executed as soon as they become available, while later actions continue to be refined. At each sub-chunk boundary, the future latent is re-predicted from the latest observation and used to update all unexecuted actions, enabling long-horizon execution without repeated full policy inference. The resulting future-prediction error can further serve as a signal for adaptive chunking. S-WAM achieves 97.7% on LIBERO and 87.9% on LIBERO-Plus, and improves performance across multiple policy backbones and real-robot tasks. It reaches 292.7 executed actions per second, $3.62\times$ the throughput of conventional execution at comparable accuracy, while reducing time-to-first-action from 123.6 to 73.3 ms. With additional inference optimizations, throughput further increases to 642.9 actions per second.

Robots operating in real-world environments must often execute complex, multi-step bimanual tasks over long horizons rather than single, isolated actions. Current manipulation datasets provide limited support for this capability: although single-arm datasets reach hundreds of thousands of trajectories, they typically provide only one high-level instruction per episode, while existing bimanual datasets provide subtask annotations for only part of their recorded hours. We present FineART, a densely annotated bimanual manipulation dataset comprising 40,543 episodes (1,718 hours) and 533,913 subtasks across 151 tasks. We also introduce FineART-VLA, a vision-language-action policy that predicts its own next subtask to guide its actions. Mid-training on FineART's subtask annotations improves FineART-VLA's success at following spatial instructions from 32.0% to 100.0%. With step-by-step human subtask guidance, success on unseen long-horizon tasks increases from 16.0% to 76.0%. Furthermore, FineART-VLA matches baseline performance on a new robot with 10x less fine-tuning data and generalizes zero-shot to unseen tasks. We open-source the full dataset, model weights, and training code.

Vision-language-action (VLA) models perform well on shorter-horizon manipulation tasks but still struggle with long-horizon tasks that require multiple dependent manipulations from a single command. Online reinforcement learning (RL) can improve these policies through environment interaction, yet many existing methods provide reward only after the complete task succeeds. However, such terminal supervision is sparse and does not distinguish early failures from rollouts that make substantial partial progress. We propose StructRL, an online RL framework that constructs structured intermediate supervision from verifiable subtask completions. StructRL decomposes each task into verifiable subtasks, grants intermediate rewards only after the prerequisite subtasks have been completed, and scales each reward according to completion pace. Across RoboCasa365 and LIBERO-Long with GR00T-N1.5 and pi 0.5, StructRL consistently outperforms evaluated online RL baselines. These results show that verifiable, structured intermediate rewards improve long-horizon VLA post-training. Code is available at https://github.com/amazon-science/StructRL.

Learning from human videos offers a promising route to acquiring diverse manipulation skills. Extending this capability beyond tabletop settings to long-horizon mobile manipulation requires adapting and composing demonstrated interactions across changing scenes and robot configurations. We present Skill Assembly and Kinematic Imitation (SAKI), a framework connecting human-video skill acquisition, cross-demonstration assembly and closed-loop whole-body execution. SAKI prepares reusable object-centric skills that preserve task-critical interactions while allowing transfer paths to adapt. Given a goal and supplied task dependencies, it selects and orders skills, binds their object roles to the current scene, and carries scene estimates and robot configuration between successive skills. Whole-body kinematic imitation generates coordinated base, arm and gripper motion. During execution, persistent object estimates maintain task references across viewpoint changes, while visual feedback updates remaining trajectories. Real-robot experiments demonstrate skill reuse across layouts and the composition of independently demonstrated interactions into continuous mobile tasks, including tidying and wiping. Ablation results show that task-conditioned reference preparation substantially improves long-horizon task completion with whole-body optimisation and visual feedback held fixed. Check https://aus.bot/research/saki/ for video demos!

Human videos offer a scalable source of demonstrations for dexterous robot manipulation. However, existing human-to-simulation-to-robot (Human2Sim2Robot) pipelines rely on predefined procedures that struggle to accommodate diverse object properties and interactions, particularly those involving articulated and deformable objects. We introduce DexAgent, an agentic Human2Sim2Robot framework that converts a single egocentric human video and a task prompt into physically grounded robot trajectories for policy training. It operates through four stages: semantic understanding of human videos, property-based simulation reconstruction, robot trajectory optimization, and robot data generation. At each stage, DexAgent adapts its approach to the task and object properties by selecting suitable skills from its tool library or developing new ones when needed. Property-specific verifiers assess stage outcomes for physical validity and task-specific requirements and provide feedback for refinement, preventing error propagation through the workflow. This adaptive, verification-guided process allows DexAgent to process diverse objects and long-horizon tasks. In the final stage, DexAgent varies object and robot states in simulation to generate diverse robot trajectories from a single human video, then retextures the rendered observations to facilitate sim-to-real transfer. Newly developed skills and verifiers are retained in its tool library, making it self-evolving to accumulate reusable capabilities. This reduces processing time as DexAgent encounters more human videos. Across eleven real-world tasks, policies trained with DexAgent-generated data achieve a 3.5x higher success rate than competing baselines. Project website: https://dexagent123.github.io/.

Pretrained robot manipulation policies such as vision-language-action models (VLAs) or world-action models (WAMs) leave interaction-relevant metric geometry implicit. Recent breakthroughs in spatial reconstruction can supply the necessary geometry reliably, but their features describe local shape without stating where it lies with respect to the robot. How best to deliver these features to a pretrained policy remains unresolved. We propose Spatial Grafting, a versatile, lightweight spatial module that binds frozen reconstruction features to metric, robot-relative geometry. Spatial Grafting constructs metric-grounded spatial tokens and injects them into the flow-matching action expert through cross-attention, without modifying the host's perceptual pathway, so the host retains the full benefit of its pretraining. We evaluate it more broadly than any geometry-aware policy we compare against: one graft architecture, with no per-host redesign, on two VLAs and two WAMs, across four simulation benchmarks that span short-horizon manipulation, visual robustness, clutter and long-horizon mobile manipulation, and on three real-robot platforms with single- and dual-arm configurations. On RoboTwin 2.0, a dual-arm manipulation benchmark, the graft improves every host across VLAs and WAMs. Grafted $π_{0.5}$ gains 11.3% and 15.6% on clean and randomized scenes, reaching 94.0% and 92.4%, above the strongest published 3D-conditioned policy, WAM4D (93.8% and 89.9%). The margin widens as the horizon lengthens: on tasks from BEHAVIOR-1K, a dual-arm mobile manipulation challenge scored by average task progress, it surpasses the 2025 challenge winner on five of six tasks,by up to 0.47 Q-score, and exceeds a map-conditioned spatial policy on average across the three tasks both report.

Long-horizon manipulation requires robots to remember cues that are no longer in view while responding to moving objects. Yet vision-language-action (VLA) policies often rely on the latest observation, and refreshing their visual context typically requires another costly vision-language model (VLM) pass. We present D$^2$-VLA, which combines dual memory and dual-frequency control at the KV-cache interface of a pretrained VLA. D$^2$-VLA uses block-wise causal KV caching to encode observations incrementally and, guided by distinct temporal attention patterns, constructs separate historical KV read views for the VLM and action expert. Between periodic VLM updates, a gated adapter incorporates fresh visual features into the latest history-conditioned KV block, while a short fast-memory queue supports action replanning. We introduce DOMINO-Long, a ten-task benchmark requiring robots to use earlier visual cues when manipulating moving objects. D$^2$-VLA achieves complete-task success rates of 29.3% on DOMINO, compared with 9.6% for $π_{0.5}$ and 17.2% for PUMA, and 60.0% on DOMINO-Long, compared with 35.4% and 20.6%, respectively. It improves success rates on eight real-robot tasks and reaches 97.5% on LIBERO-Long and 74.3% on RoboTwin 2.0.

Memory is essential for long-horizon, partially observed robotic manipulation: a robot must remember which object was placed in a drawer, whose cup it moved, or how many action cycles have elapsed. Recent vision-language-action (VLA) models embed memory directly inside the policy, but benchmarks show no single in-policy mechanism covers all spatio-temporal dimensions, trailing oracle methods by a wide margin. We argue that memory type dictates where memory should reside: short-term perceptual memory (repetition, timing, retracing) belongs inside the policy, while long-term object memory (persistent spatial state, containment, event history) belongs outside as an explicit, readable record. We present Ledger, a harness that realizes this split over a single fine-tuned $π_{0.5}$ policy by pairing an in-policy frame-sampling memory with an external spatio-temporal object memory, the ledger, built from a SAM3 tracker and a VLM captioner of the demonstration and read by an LLM planner that decides at step boundaries. On RoboMME, Ledger reaches the highest four-suite average among the evaluated methods, 64.3% (vs. 45.9% for the strongest prior method under identical evaluation), leading object reference (60.7% vs. 40.3%) and object permanence (86.7% vs. 56.2%) using a single set of weights. Choosing the memory source at runtime, from the instruction and the record, removes the need for a task-level router.

General-purpose vision-language models offer a promising way to zero-shot robot control: \gptastra{} excels at open-ended and language- or image-conditioned manipulation but remains substantially weaker on high-precision and long-horizon tasks. We introduce \emph{RoboICL}, an in-context robot-control framework that narrows these gaps without robot-specific parameter updates or a learned VLA. RoboICL separates \emph{demonstration context}, which provides recorded examples when available, from \emph{interaction memory}, which accumulates the model's own actions and observed outcomes. Both use a shared observation--action--receipt--observation grammar. To preserve experience across task stages, RoboICL combines sampled demonstration blocks with bounded anchored memory. Fixed anchors keep earlier rollout interactions available for in-context learning, while the latest interaction supports immediate error correction. Across 30 RoboDojo tasks, using zero shot for Open and one demonstration elsewhere, RoboICL improves on official zero-shot \gptastra{} by 20--27 progress-score points in every category. It leads the leaderboard baselines on Memory and Open, achieves comparable performance to the strongest Precision baseline, and remains competitive on Long-Horizon. Its 30-task Overall score is 50.64, versus 33.68 for the strongest baseline. On a separate ten-task subset, RoboICL scores 60.60, within 2.00 points of the $π_{0.5}$ + \gptastra{} hybrid approach. On three real-robot tasks, mean progress rises from 14.45 at zero shot to 63.33 at one shot and 78.89 at three shots. On two development tasks, optional Jev-gated action reuse reduces \gptastra{} calls by 33--48%. Code is available at https://github.com/Mosi-AI/RoboICL.

Robotic manipulation policies are advancing rapidly with increasing reliance on vision-language models for end-to-end decision making. However, reliable deployment remains challenging because many policies lack explicit mechanisms for predicting task outcomes and evaluating whether generated actions will achieve desired final states, causing execution errors to accumulate during long-horizon manipulation. We present Robot-GST, a geometry-aware spatio-temporal behaviour representation and evaluation framework that constructs a Gaussian-SAM robotic environment for real-to-sim policy verification and improves the reliability of real-world manipulation deployment. Our approach constructs a high-fidelity robotic environment from RGB-D observations using 3D Gaussian Splatting and SAM3D, enabling ``simulation and evaluation before acting''. It integrates visual observations and language instructions with spatio-temporal reasoning for long-horizon task planning using large vision-language models. To bridge high-level planning and real-world execution, we introduce Gaussian-aware final-state estimation through geometric sampling and state-based trajectory planning. Before execution, candidate action sequences are simulated and evaluated in the Gaussian-SAM environment to filter infeasible behaviours. We validate our approach on representative manipulation tasks involving rigid, soft, and deformable objects, including cube placing, toy packing, and duck rearrangement, demonstrating that geometry-aware spatio-temporal reasoning and state-aware execution improve manipulation reliability across different object categories. Our results suggest that combining geometry-aware reconstruction with high-quality rendering and simulation provides a scalable approach for evaluating robotic manipulation behaviours. Website: https://robot-gst.github.io