Long-Horizon Robotic Manipulation

Generalist embodied agents must perform interactive, causally-dependent reasoning, continually interacting with the environment, acquiring information, and updating plans to solve long-horizon tasks before they could be adopted in real-life scenarios. For instance, retrieving an apple from a cabinet may require opening multiple doors and drawers before the apple becomes visible and reachable, demanding sequential interaction under partial observability. However, existing benchmarks fail to systematically evaluate this essential capability. We introduce COIN, a benchmark designed to assess interactive reasoning in realistic robotic manipulation through three key contributions. First, we construct COIN-50: 50 interactive tasks in daily scenarios, and create COIN-Primitive required by causally-dependent tasks, and COIN-Composition with mid-term complexity for skill learning and generalization evaluation. Second, we develop a low-cost mobile AR teleoperation system and collect the COIN-Primitive Dataset with 50 demonstrations per primitive task (1,000 in total). Third, we develop systematic evaluation metrics about execution stability and generalization robustness to evaluate CodeAsPolicy, VLA, and language-conditioned H-VLA approaches. Our comprehensive evaluation reveals critical limitations in current methods: models struggle with interactive reasoning tasks due to significant gaps between visual understanding and motor execution. We provide fine-grained analysis of these limitations.

Robotic manipulation policies often degrade over extended horizons, yet existing benchmarks provide limited insight into why such failures occur. Most prior benchmarks are either simulation-based or report aggregate success, making it difficult to disentangle the distinct sources of temporal difficulty in real-world execution. We introduce LongBench, a real-world benchmark for evaluating long-horizon manipulation. LongBench consists of over 1,000 real-world episodes, covering two complementary regimes: Context-Independent (fully observable) and Context-Dependent (ambiguity-driven). By organizing tasks into capability- and ambiguity-specific subsets, LongBench enables mechanism-aware evaluation of execution robustness, temporal consistency, and context-dependent reasoning. Evaluating six state-of-the-art policies reveals that long-horizon performance is not governed by a single factor. We observe that performance in fully observable settings is more strongly associated with execution robustness, while contextual difficulty varies across tasks and is not consistently improved by memory-based methods. We hope that LongBench serves as a useful benchmark for studying long-horizon manipulation and for developing policies with stronger robustness across both execution and contextual challenges.

Vision-Language-Action (VLA) models have demonstrated significant potential for embodied decision-making; however, their application in complex chemical laboratory automation remains restricted by limited long-horizon reasoning and the absence of persistent experience accumulation. Existing frameworks typically treat planning and execution as decoupled processes, often failing to consolidate successful strategies, which results in inefficient trial-and-error in multi-stage protocols. In this paper, we propose ChemBot, a dual-layer, closed-loop framework that integrates an autonomous AI agent with a progress-aware VLA model (Skill-VLA) for hierarchical task decomposition and execution. ChemBot utilizes a dual-layer memory architecture to consolidate successful trajectories into retrievable assets, while a Model Context Protocol (MCP) server facilitates efficient sub-agent and tool orchestration. To address the inherent limitations of VLA models, we further implement a future-state-based asynchronous inference mechanism to mitigate trajectory discontinuities. Extensive experiments on collaborative robots demonstrate that ChemBot achieves superior operational safety, precision, and task success rates compared to existing VLA baselines in complex, long-horizon chemical experimentation.

Existing robotic foundation policies are trained primarily via large-scale imitation learning. While such models demonstrate strong capabilities, they often struggle with long-horizon tasks due to distribution shift and error accumulation. While reinforcement learning (RL) can finetune these models, it cannot work well across diverse tasks without manual reward engineering. We propose VLLR, a dense reward framework combining (1) an extrinsic reward from Large Language Models (LLMs) and Vision-Language Models (VLMs) for task progress recognition, and (2) an intrinsic reward based on policy self-certainty. VLLR uses LLMs to decompose tasks into verifiable subtasks and then VLMs to estimate progress to initialize the value function for a brief warm-up phase, avoiding prohibitive inference cost during full training; and self-certainty provides per-step intrinsic guidance throughout PPO finetuning. Ablation studies reveal complementary benefits: VLM-based value initialization primarily improves task completion efficiency, while self-certainty primarily enhances success rates, particularly on out-of-distribution tasks. On the CHORES benchmark covering mobile manipulation and navigation, VLLR achieves up to 56% absolute success rate gains over the pretrained policy, up to 5% gains over state-of-the-art RL finetuning methods on in-distribution tasks, and up to $10\%$ gains on out-of-distribution tasks, all without manual reward engineering. Additional visualizations can be found in https://silongyong.github.io/vllr_project_page/

Reinforcement Learning (RL) has shown strong potential for improving robotic manipulation policies, yet its practical use remains bottlenecked by the difficulty of specifying reward functions that are both semantically meaningful and reusable across tasks. In this paper, we propose Large Reward Models (LRMs), a framework that adapts foundation VLMs into frame-level reward generators for robot policy refinement. We specialize a state-of-the-art VLM on a multi-source dataset spanning real-world robot trajectories, human-object interactions, and simulated manipulation environments. Unlike prior approaches that mainly evaluate trajectories post-hoc, LRMs expose multiple reward interfaces from visual observations: progress estimation, task completion, and temporal contrastive comparison. Starting from an imitation-learned policy, we use these VLM-derived rewards to guide PPO refinement on held-out long-horizon manipulation tasks. Our experiments show that LRM progress rewards provide the strongest non-privileged online refinement signal, improving the IL baseline and narrowing the gap to privileged environment rewards. We further deploy progress rewards for progress-weighted behavioral cloning on four real-world manipulation tasks spanning two robot platforms, improving over SFT on all four tasks. These results suggest that modality-specific specialization of foundation VLMs can provide practical visual reward signals for both simulated policy refinement and physical robot self-improvement without hand-coded task rewards.

Learning generalizable and robust behavior cloning policies requires large volumes of high-quality robotics data. While human demonstrations (e.g., through teleoperation) serve as the standard source for expert behaviors, acquiring such data at scale in the real world is prohibitively expensive. This paper introduces ExpertGen, a framework that automates expert policy learning in simulation to enable scalable sim-to-real transfer. ExpertGen first initializes a behavior prior using a diffusion policy trained on imperfect demonstrations, which may be synthesized by large language models or provided by humans. Reinforcement learning is then used to steer this prior toward high task success by optimizing the diffusion model's initial noise while keep original policy frozen. By keeping the pretrained diffusion policy frozen, ExpertGen regularizes exploration to remain within safe, human-like behavior manifolds, while also enabling effective learning with only sparse rewards. Empirical evaluations on challenging manipulation benchmarks demonstrate that ExpertGen reliably produces high-quality expert policies with no reward engineering. On industrial assembly tasks, ExpertGen achieves a 90.5% overall success rate, while on long-horizon manipulation tasks it attains 85% overall success, outperforming all baseline methods. The resulting policies exhibit dexterous control and remain robust across diverse initial configurations and failure states. To validate sim-to-real transfer, the learned state-based expert policies are further distilled into visuomotor policies via DAgger and successfully deployed on real robotic hardware.

Long-horizon indoor mobile manipulation (MoMa) requires robots to execute extended navigation-manipulation sequences whose feasibility depends on state changes induced by preceding skills. Vision-language models (VLMs) can decompose instructions into plausible skill sequences, but they do not reliably track such cumulative embodiment constraints or revise a plan when execution deviates from expectation. We present MoMaStage, a map-light framework for state-consistent planning and closed-loop execution in long-horizon indoor MoMa. MoMaStage couples a frozen VLM with robot execution through three mechanisms: (i) a hierarchical library of grounded, executable skills and a topology-only projection of a Skill-State Graph (SSG) that constrains the VLM's planning space; (ii) an SSG verifier that propagates scene-region and gripper-occupancy state to reject infeasible plans before execution; and (iii) an event-driven monitor that triggers graph-grounded repair only when an observed outcome invalidates the remaining plan. The SSG captures the compact embodiment state needed for skill sequencing without requiring a dense scene map, while geometric and contact-level conditions remain within the underlying controllers. Experiments in physics-rich simulation and on a real mobile manipulator show that MoMaStage improves planning validity and long-horizon execution survival over the evaluated baselines, while reducing latency and model token consumption.

Vision-language-action (VLA) models integrate visual observations and language instructions to predict robot actions, demonstrating promising generalization in manipulation tasks. However, most existing approaches primarily rely on direct mappings from 2D visual inputs to action sequences, without explicitly modeling the underlying 3D spatial structure or temporal world dynamics. Such representations may limit spatial reasoning and long-horizon decision-making in dynamic environments. To address this limitation, we propose StemVLA, a novel framework that explicitly incorporates both future-oriented 3D spatial knowledge and historical 4D spatiotemporal representations into action prediction. First, instead of relying solely on observed images, StemVLA forecasts structured 3D future spatial-geometric world knowledge, enabling the model to anticipate upcoming scene geometry and object configurations. Second, to capture temporal consistency and motion dynamics, we feed historical image frames into a pretrained video-geometry transformer backbone to extract implicit 3D world representations, and further aggregate them across time using a temporal attention module, termed VideoFormer [20], forming a unified 4D historical spatiotemporal representation. By jointly modeling 2D observations, predicted 3D future structure, and aggregated 4D temporal dynamics, StemVLA enables more comprehensive world understanding for robot manipulation. Extensive experiments in simulation demonstrate that Stem-VLA achieves an average accuracy of 92.0% across the LIBERO subsets, and 86.0% on the long-horizon LIBERO-Long subset.

Solving complex long-horizon robotic tasks requires joint reasoning over abstract task structure and low-level physical interaction. While combining Vision-Language Models (VLMs) and video generation models offers a promising path for zero-shot planning, their individual tendencies to hallucinate physics or violate geometric consistency often compound over time, preventing reliable real-world execution. We introduce NovaPlan, a hierarchical framework that enables robust, zero-shot long-horizon manipulation by systematically proposing, verifying, and repairing visual plans. At the high level, a VLM planner decomposes tasks and filters out dynamically inconsistent futures by verifying multiple candidate video rollouts. To translate these imagined futures into reliable physical actions, NovaPlan utilizes a hybrid geometric representation that adaptively switches between object-centric flow and human hand flow. Finally, NovaPlan closes the loop by continuously monitoring execution to verify outcomes and synthesize local, non-prehensile corrective behaviors, such as fingertip poking, when failures occur. Across diverse multi-stage tasks, NovaPlan substantially outperforms prior zero-shot systems, achieving complex assembly and dexterous error recovery entirely without task-specific training or demonstrations. Please visit our project website for additional results: https://nova-plan.github.io/

Robot navigation typically assumes an obstacle-free path exists between start and goal. In real environments, however, clutter may block all routes. We introduce Lifelong Interactive Navigation, where a mobile robot with manipulation capabilities must move objects to forge paths and complete sequential object-placement tasks. Because environment modifications persist, decisions impact future navigability and task difficulty. We propose CoReLIN, an LLM-driven constraint-based reasoning framework with active perception. CoReLIN reasons over a structured scene graph to decide which objects to relocate, where to place them, and where to explore next. A standard motion planner executes reliable navigation and manipulation primitives. To evaluate long-horizon behavior, we introduce 2 new metrics - Long-term Efficiency Score (LES), a unified metric capturing success, execution efficiency, environment optimality, captured by Price of Clutter. In ProcTHOR-10k, CoReLIN outperforms best baseline by 16% under standard metrics and LES, and transfers to real-world hardware.

World models are becoming central to robotic planning and control by predicting future state transitions. Existing approaches mainly rely on visual, latent, or language prediction, which can be difficult to ground in executable robot actions and prone to compounding errors over long horizons. In contrast, traditional robotic task and motion planning enables structured long-horizon reasoning through compact symbolic representations of world transitions, but typically lacks synchronized visual prediction. We propose Hierarchical World Model (H-WM), which jointly predicts logical and visual state transitions by combining a high-level logical world model with a low-level visual world model. The predicted logical actions and latent visual state transitions are jointly incorporated into Vision-Language-Action (VLA) models as intermediate state guidance for long-horizon task execution. Experiments on three long-horizon benchmarks and real robots show that H-WM consistently improves VLA's performance by stabilizing long-horizon execution and mitigating error accumulation. We also construct LIBERO-Logic, a frame-level aligned dataset that pairs visual observations and continuous robot states with logical actions and predicate-based logical states.

Learning long-horizon robotic manipulation requires jointly achieving expressive behavior modeling, real-time inference, and stable execution, which remains challenging for existing generative policies. Diffusion-based approaches offer strong modeling capacity but incur high inference latency, while flow matching enables fast, near-single-step generation yet often suffers from unstable execution when operating directly in the raw action space. We propose Continuous Latent Action Flow Policy (CoLA-Flow Policy), a trajectory-level imitation learning framework that performs flow matching in a continuous latent action space. By encoding action sequences into temporally coherent latent trajectories and learning an explicit latent-space flow, CoLA-Flow Policy decouples global motion structure from low-level control noise, enabling smooth and reliable long-horizon execution. The framework further integrates geometry-aware point cloud conditioning and execution-time multimodal modulation, using visual cues as a representative modality to enhance real-world robustness. Experiments in simulation and on real robots show that CoLA-Flow Policy achieves near-single-step inference, improves trajectory smoothness by up to 93.7% and task success by up to 25 percentage points over raw action-space flow baselines, while remaining significantly faster than diffusion-based policies.

Scaling imitation learning to diverse multi-task robot manipulation remains challenging due to suboptimal demonstrations, behavioral multi-modality, and destructive interference across tasks. While skill-based methods offer a promising direction by decomposing behaviors into reusable abstractions, existing approaches often learn skills that are either biased toward linguistic structure or lack semantic alignment across tasks, limiting generalization. In this work, we propose AtomSkill, a novel framework that learns a semantically aligned Atomic Skill Space from demonstrations and enables robust long-horizon execution through keypose imagination. Our method introduces: (1) semantic contrastive skill alignment, which partitions demonstrations into variable-length atomic skills and employs a contrastive objective to jointly enforce semantic consistency and temporal coherence, yielding a compact and reusable skill library; and (2) action decoding with keypose imagining, where the policy predicts both a skill's terminal keypose and immediate actions, thereby supporting progress-aware skill transitions. During inference, an atomic skill diffusion sampler generates plausible skill sequences, while predicted keyposes autonomously trigger smooth skill chaining. Extensive experiments in simulation and real-world settings show that AtomSkill consistently outperforms state-of-the-art imitation learning and skill-based baselines. Project page: https://atom-skill.github.io.

Scalable embodied intelligence is constrained by the scarcity of diverse, long-horizon robotic manipulation data. Existing video world models in this domain are limited to synthesizing short clips of simple actions and often rely on manually defined trajectories. To this end, we introduce MIND-V, a cognitive hierarchical world model designed to synthesize physically plausible and logically coherent videos of long-horizon robotic manipulation. Inspired by cognitive science, MIND-V bridges high-level reasoning with pixel-level synthesis through three core components: a Semantic Reasoning Hub (SRH) that leverages a pre-trained vision-language model for task planning; a Behavioral Semantic Bridge (BSB) that translates abstract instructions into domain-invariant representations; and a Motor Video Generator (MVG) for conditional video rendering. MIND-V employs Staged Visual Future Rollouts, a test-time optimization strategy to enhance long-horizon robustness. To enforce adherence to physical laws, we introduce a GRPO reinforcement learning post-training phase guided by a novel Physical Foresight Coherence (PFC) reward. PFC leverages the V-JEPA2 world model as a physics referee to penalize implausible dynamics in the latent feature space. Experiments confirm MIND-V's SOTA performance in long-horizon simulation and its significant value for policy learning, introducing a scalable and fully autonomous framework for embodied data synthesis.

Building robots that can perceive, reason, and act in dynamic, unstructured environments remains a central challenge. Recent embodied systems often follow a dual-system paradigm, where System 2 performs high-level reasoning and System 1 handles low-level control. We refer to System 2 as the embodied brain, the cognitive core for decision-making in manipulation. Although evaluating this embodied brain is crucial, existing benchmarks mainly measure execution success or cover only limited aspects of high-level cognition and task realism. We introduce RoboBench, a benchmark for evaluating multimodal large language models (MLLMs) as embodied brains. RoboBench covers five dimensions: Instruction Comprehension, Perception Reasoning, Generalized Planning, Affordance Prediction, and Failure Analysis. It spans 14 capabilities, 25 tasks, and 6,092 QA pairs. To improve realism, it draws from large-scale real robotic data and in-house collection across diverse embodiments, attribute-rich objects, multi-view scenes, and memory-driven navigation. For planning, RoboBench introduces an MLLM-as-world-simulator framework that assesses whether predicted plans can achieve critical object-state changes under physical and visual constraints, enabling more faithful evaluation of long-horizon reasoning than symbolic matching. Experiments on 18 state-of-the-art MLLMs reveal persistent limitations in implicit instruction understanding, spatiotemporal reasoning, cross-scenario planning, fine-grained affordance understanding, and failure diagnosis. We further analyze how embodied cognitive abilities relate to downstream robotic control. RoboBench offers a comprehensive scaffold for quantifying high-level cognition and guiding next-generation MLLMs toward more robust robotic intelligence.

Despite rapid progress in robotics, complex or long-horizon tasks remain a fundamental challenge. Most current approaches follow an open-loop paradigm with limited reasoning and no feedback, resulting in poor robustness to environmental changes and severe error accumulation. We present RoboPilot, a dual-thinking closed-loop agentic framework for robotic manipulation that supports adaptive reasoning for complex tasks in real-world dynamic environments. RoboPilot leverages primitive actions for structured task planning and flexible action generation as a agentic system, while introducing feedback to enable replanning from dynamic changes and execution errors. Chain-of-Thought reasoning further enhances high-level task planning and guides low-level action generation. The agentic system dynamically switches between fast and slow thinking to balance efficiency and accuracy. To systematically evaluate the robustness of RoboPilot in diverse robot manipulation scenarios, we introduce RoboPilot-Bench, a benchmark spanning 21 tasks across 10 categories, including infeasible-task recognition and dynamic recovery. Experiments show that RoboPilot outperforms state-of-the-art baselines by 11% in task success rate, and the real-world deployment on an industrial robot further demonstrates its robustness.

Large language models (LLMs) can translate natural-language instructions for robotic manipulation into executable code, but ambiguity, noisy generations, and limited context windows make ultra-long-horizon tasks unreliable. Closed-loop approaches that rely only on LLM feedback also struggle because LLMs have limited robotic reasoning, even when task errors are obvious to humans. Feedback is often stored in representations that generalize poorly to unseen tasks and can cause catastrophic forgetting as new corrections accumulate. We propose LYRA, a human-guided lifelong skill learning and code generation framework that distills human feedback into modular, reusable skills and incrementally extends their functionality across successive interactions while preserving previously learned behavior. External memory stores learned skills and execution examples; retrieval-augmented generation selects relevant knowledge, while user hints guide reuse when retrieval is insufficient, supporting ultra-long-horizon execution. Experiments on Ravens, Franka Kitchen, LIBERO-long, MetaWorld, and real-world tasks show a 0.93 success rate, up to 27% higher than baselines, and a 42% improvement in correction efficiency. LYRA also robustly solves ``build a house'', which requires planning over 20 primitives.

Humanoid robots have seen remarkable advances in dexterity, balance, and locomotion, yet their role in expressive domains such as music performance remains largely unexplored. Musical tasks, like drumming, present unique challenges such as split-second timing, rapid contacts, and multi-limb coordination over performances lasting minutes. In this paper, we introduce Robot Drummer, a simulation framework for humanoid drumming across a diverse repertoire of songs. We formulate humanoid drumming as the realization of timed contact events encoded as a Rhythmic Contact Chain. To handle the long-horizon nature of musical performance, we decompose each track into fixed-length segments and train a single policy across all segments in parallel using reinforcement learning. Through extensive experiments on over thirty popular tracks, our results demonstrate that Robot Drummer consistently achieves high F1 scores and enables efficient learning of long-horizon musical performances. The learned behaviors exhibit emergent human-like drumming strategies, such as cross-arm strikes, and adaptive stick assignments, demonstrating the potential of reinforcement learning to bring humanoid robots into the domain of creative musical performance. Project page: robotdrummer.github.io

Vision-language-action (VLA) models have shown promising progress in robotic manipulation. However, directly mapping visual observations and language instructions to low-level actions often results in limited interpretability and weak robustness in complex, long-horizon tasks. To address these challenges, we employ a modular manipulation framework that separates high-level planning from low-level control. At its core is Gondola, a grounded vision-language planning model that generates structured plans with explicit pixel-level object grounding before action execution. Given multi-view observations and planning history, Gondola predicts the next-step plan as interleaved textual instructions and multi-view segmentation masks corresponding to target objects and goal locations. To train Gondola, we construct synthetic datasets that provide explicit supervision for short-horizon grounded planning, multi-view referring expression, and long-horizon compositional reasoning. By coupling grounded plan generation with a 3D-based execution policy, our framework achieves state-of-the-art performance on the challenging GemBench benchmark. The system further demonstrates promising transfer to real robots. Ablation studies confirm that pixel-level grounding and the proposed planning-oriented supervision are critical for effective high-level reasoning. Project webpage: https://cshizhe.github.io/projects/robot_gondola.html

Planning long-horizon manipulation motions using a set of predefined skills is a central challenge in robotics; solving it efficiently could enable general-purpose robots to tackle novel tasks by flexibly composing generic skills. Solutions to this problem lie in an infinitely vast space of parameterized skill sequences -- a space where common incremental methods struggle to find sequences that have non-obvious intermediate steps. Some approaches reason over lower-dimensional, symbolic spaces, which are more tractable to explore but may be brittle and are laborious to construct. In this work, we introduce MOSAIC, a skill-centric, multi-directional planning approach that targets these challenges by reasoning about which skills to employ and where they are most likely to succeed, by utilizing physics simulation to estimate skill execution outcomes. Specifically, MOSAIC employs two complementary skill families: Generators, which identify ``islands of competence'' where skills are demonstrably effective, and Connectors, which link these skill-trajectories by solving boundary value problems. By focusing planning efforts on regions of high competence, MOSAIC efficiently discovers physically-grounded solutions. We demonstrate its efficacy on complex long-horizon problems in both simulation and the real world, using a diverse set of skills including generative diffusion models, motion planning algorithms, and manipulation-specific models. Visit skill-mosaic.github.io for demonstrations and examples.

Building long-horizon robot agents requires composing closed-loop pipelines -- perception, belief update, planning, and control -- whose components run at different clocks and with variable latency. Today, these systems are often assembled with ad-hoc concurrency and pub/sub conventions that make timing and input-consumption semantics implicit, yielding schedule-dependent behavior that is hard to reproduce, debug, and reuse. Current solutions typically solve parts of this problem at either the algorithmic or the systems layer, but not both. In this work, we propose Retriever, which spans the entire stack: an asynchronous decision model, a programming model, a runtime, and an example closed-loop agent pipeline. Retriever represents an agent as a graph of stateful causal stream functions executed on explicit run clocks. We formalize this view via an asynchronous environment-agent loop over continuous-time streams and show that finite-memory causal policies can be represented by compositions of these operators. Retriever compiles these graphs into a runtime that supports multiple backends, enabling systematic debugging across running environments and deterministic replay from logged asynchronous data. We evaluate Retriever through a real-robot case study together with controlled studies of runtime overhead and deterministic replay behavior.