Papers

HIL-UMI: Bringing Human-in-the-Loop Post-Training of Vision-Language-Action Models to Universal Manipulation Interface

Large-scale vision-language-action (VLA) models provide powerful priors for robot manipulation, yet adapting them to a specific deployment remains challenging. Supervised fine-tuning (SFT) on task-specific demonstrations provides a step toward deployment, but faces two persistent limitations: static data provide limited coverage of out-of-distribution states, and standard imitation objectives do not distinguish progressing behavior from less useful data. Interactive post-training can address these limitations, but typically requires repeated policy execution and human intervention on a physical robot. We introduce HIL-UMI, a policy-guided Universal Manipulation Interface (UMI) framework for robot-free human-in-the-loop VLA post-training. During handheld UMI demonstrations, HIL-UMI queries the current policy on the same observation stream without executing its predictions. The Energy Score compares the human action trajectory with policy inference and triggers collection when their discrepancy indicates an out-of-distribution region. In a separate feedback loop, low online advantage predictions identify essential segments for refining a progress-based advantage estimator. The updated estimator then guides advantage-conditioned behavioral cloning using a balanced mixture of base demonstrations and new policy data. This design preserves the iterative and policy-aware nature of human-in-the-loop learning while decoupling data collection from robot deployment. Experiments on four real-world tasks spanning long-horizon and precise manipulation show that HIL-UMI achieves consistent improvement over SFT and benefits from both targeted collection and advantage refinement. Moreover, HIL-UMI outperforms HG-DAgger on Clean Up Table with lower per-frame collection time, suggesting a scalable path for VLA post-training across operators and locations.

VLAHuman-in-the-LoopPost-trainingHan, Zimu, Zeng, Yiming, Zhang, Jiyao·Sep 17, 2026

CAP: Continuously Adaptive Perception-Blind Humanoid Locomotion via Learned Denoising

Humanoid locomotion across complex terrain demands forward-looking exteroception to anticipate obstacles, yet this signal is unreliable in real-world deployment, failing partially and intermittently. Existing perceptive policies often assume that depth observations remain clean and in-distribution, while recent attempts to unify perceptive and blind control typically route or switch between separate sub-policies, leaving recoverable information in partially corrupted depth unexploited. We instead propose CAP, a single-stage humanoid locomotion policy that recovers this signal with a perceptive world-model encoder trained as a learned denoiser to reconstruct clean depth from a corrupted input, together with a co-active proprioceptive variational encoder that supplies depth-free body-state information. A coupled training recipe pairs a depth-noise curriculum on the world-model input with world-model feature dropout on the policy-facing latent, exposing the policy to failures across the entire perception-quality spectrum. In simulation, CAP matches or improves upon perceptive baselines when depth remains informative, and degrades more smoothly than a binary-switching baseline as perception worsens. On the Unitree G1, controlled trials and indoor-outdoor deployments demonstrate perception-robust locomotion under intermittent occlusion, real-sensor corruption, and outdoor depth artifacts.

Humanoid LocomotionWorld ModelsDepth DenoisingChen, Hongjin, Xu, Zijun, Ma, Shihao·Sep 10, 2026

WholeBodyWAM: Generalizing Pre-trained World-Action Priors to Humanoid Loco-Manipulation via WBC-Grounded Coordination

World Action Models (WAMs) offer a promising approach to general-purpose robot manipulation by jointly modeling visual dynamics and actions. However, most WAM studies focus on tabletop or arm-centric manipulation, while humanoid loco-manipulation remains less explored. To address this gap, we introduce WholeBodyWAM, which jointly predicts future visual dynamics, manipulation actions, and whole-body control intents for generalizable humanoid loco-manipulation. It preserves pre-trained world-action priors while grounding heterogeneous whole-body controller (WBC) semantics and coordinating whole-body behavior. Extensive experiments show that WholeBodyWAM achieves an overall simulation task success rate of 91.9%, with a 0.23 improvement in real-world out-of-distribution task progress and a 70% reduction in success-rate variance across WBCs relative to the respective baselines. These results suggest a path toward scalable humanoid whole-body intelligence by extending pre-trained world-action priors through structured WBC grounding and coordination, rather than relearning whole-body behavior from scratch. Project page: https://wholebodywam.github.io/.

World-action modelsHumanoidMobile ManipulationLi, Zhuo, Yao, Yiming, Tan, Jim·Sep 15, 2026

JEPA-Anything: Learning Predictive Models across Different Worlds

JEPA-Anything introduces orthogonal predictive factorization for domain-agnostic world modeling, splitting latent targets into complementary factors with dedicated predictive pathways. Across seven domains, it improves matched dynamics tasks and supports intervention, OOD generalization, and long-horizon forecasting.

World ModelsJEPAPredictive Representation LearningCui, Taoyong, Wang, Zhongyao, Xu, Xinyue·Sep 17, 2026

SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness

As coding agents move from supervised code completion to unattended, around-the-clock exploration, their work expands into long trajectories of reasoning, tool use, and feedback. SoL-Pi scales recursive auto-research across executable environments to discover reusable harness mechanisms. Four retained mechanisms improve action execution, context compaction, observation handling, and delegated reading, reducing recorded token traffic by 44.7-49.0% and API cost by about one third at comparable EdgeBench performance.

Agent HarnessRecursive self-improvementToken EfficiencyHaozhe Liu, Tian Ye, Sensen Gao·Sep 17, 2026

ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement

ModularRSI is a benchmark-disjoint, contrastive, and modular framework for generalizable agent-harness evolution. It diagnoses recurring failures from paired successful and failed trajectories, evolves five functional modules independently, and integrates validated changes into a unified harness. On Terminal-Bench 2.0 and SWE-Bench Verified, the evolved harness improves unseen in-domain and cross-domain tasks and transfers across foundation models.

Agent HarnessRecursive self-improvementAI AgentSiwei Wu, Jincheng Ren, Yizhi Li·Sep 14, 2026

SSR: Scaling Surefooted and Symmetric Humanoid Traversal to the Open World

Extending humanoid traversal to the open world is key to practical deployment in human environments, but remains challenging. The robot must use vision to ensure safe and reliable foot placement on heterogeneous terrain under highly dynamic motion, while producing coordinated, natural whole-body behaviors. We propose SSR, an efficient end-to-end framework for egocentric vision-based humanoid traversal that jointly learns these capabilities. SSR introduces imagined foothold guidance, which learns to model forthcoming swing-foot contacts and evaluates their support to guide pre-touchdown swings toward stable regions, reducing edge slips. It further employs equivariant latent-space symmetry augmentation to efficiently induce bilateral coordination under high-dimensional visual observations, and uses terrain-specific multi-discriminator motion priors to encourage human-like behavior across scenes. Extensive experiments show that SSR achieves safe, stable, and high-quality locomotion on diverse real-world terrains, including stairs with varied structures and extreme challenges such as wide gaps and high platforms, while enabling reliable long-horizon traversal in open outdoor environments.

HumanoidLegged LocomotionReinforcement LearningRuiqi Yu, Yiwen Wang, Yuan Hao·May 29, 2026

X-WBC: A Cross-Embodiment Foundation Model for Humanoid Whole-Body Control

Scaling humanoid whole-body control toward general-purpose deployment requires large human motion corpora and training experience shared across robot bodies. Existing methods usually train one policy per robot, leaving motion experience isolated across embodiments. We introduce X-WBC, a cross-embodiment foundation framework that separates relatively shared human motion semantics from embodiment-specific physical execution. Human-centered command tokens align full human motion, robot reference motion, and sparse VR observations. A causal Transformer learns reusable temporal structure from mixed multi-robot rollouts, while lightweight robot-specific modules map the shared representation to each robot's proprioception and action space. Across nine simulated embodiments, external motions, and four real robots, experiments show that joint training improves tracking, the aligned representation supports consistent control across command sources, and the learned policy remains competitive beyond the training corpus. These results support heterogeneous humanoids as joint data sources and establish cross-embodiment joint training as a practical route toward whole-body control foundation models.

HumanoidWhole-Body ControlCross-Embodiment LearningJuntong Zhang, Chun Gu, Li Zhang·Sep 14, 2026

Dream-RSI: Recursive Self-Improvement through Evolving Worlds

Recursive self-improvement is becoming increasingly vital for autonomous AI agents, where progress hinges on discovering high-value solutions across complex domains. The driver of this process is effective exploration, however, managing and improving exploration strategies remains a major bottleneck. Current systems face a fundamental dilemma: fixed strategies fail to adapt as search spaces scale, while online policy optimization requires navigating vast meta-search spaces under delayed and expensive feedback over long-horizon rollouts. We introduce {Dream-RSI}, a framework for scalable and recursively self-improving exploration. A lightweight orchestration layer makes exploration explicit and programmable while leaving the underlying coding agent unchanged. Our key insight is that accumulated discovery history can serve as a replay simulator over the realized search space. By performing dreaming in the replay simulator constructed from historical discovery trees, {Dream-RSI} secures immediate, low-cost off-policy feedback to evaluate and refine exploration policies without invoking repetitive, expensive online evaluations. The improved policy is subsequently redeployed online to drive further discovery, continuously expanding the simulator pool in a self-improving loop. Across algorithm engineering, mathematical optimization, and GPU kernel engineering, {Dream-RSI} achieves competitive or improved discovery quality while substantially reducing discovery cost in several settings.

Recursive self-improvementWorld ModelsLLM AgentTong Zheng, Xidong Wu, Zheng Zhang·Sep 14, 2026

SteadyTray: Learning Object Balancing Tasks in Humanoid Tray Transport via Residual Reinforcement Learning

Stabilizing unsecured payloads against the inherent oscillations of dynamic bipedal locomotion remains a critical engineering bottleneck for humanoids in unstructured environments. To solve this, we introduce ReST-RL, a hierarchical reinforcement learning architecture that explicitly decouples locomotion from payload stabilization, evaluated via the SteadyTray benchmark. Rather than relying on monolithic end-to-end learning, our framework integrates a robust base locomotion policy with a dynamic residual module engineered to actively cancel gait-induced perturbations at the end-effector. This architectural separation ensures steady tray transport without degrading the underlying bipedal stability. In simulation, the residual design significantly outperforms end-to-end baselines in gait smoothness and orientation accuracy, achieving a 96.9% success rate in variable velocity tracking and 74.5% robustness against external force disturbances. Successfully deployed on the Unitree G1 humanoid hardware, this modular approach demonstrates highly reliable zero-shot sim-to-real generalization across various objects and external force disturbances.

HumanoidResidual Reinforcement LearningMobile ManipulationAnlun Huang, Zhenyu Wu, Soofiyan Atar·Mar 11, 2026

ReST-RL: Reinforcing LLM Reasoning through Unified Self-Training and Value-Guided Search

GRPO is the representative RL method for improving LLM reasoning, yet it only ever sees one sparse reward at the end of a whole trajectory: when the rewards inside a sampled group land close together, the group-relative advantage collapses into noise and the policy learns almost nothing. ReST-RL reconnects policy optimization and value-guided search into a single self-training pipeline. Stage one, ReST-GRPO, first filters out low-information prompts by reward standard deviation, then draws prefixes from each prompt's highest-reward trajectory under a discrete exponential distribution and uses them as fresh online-GRPO starting contexts. A selected prefix is context only; its suffix is re-sampled and optimized rather than imitated. Stage two, VM-MCTS, runs MCTS under the now-static policy to self-collect value targets and trains a value model that predicts expected terminal reward. At inference the same model both allocates the tree search through UCT and ranks completed candidates in a Best-of-N fashion, so search and verification share one state-value scale. On coding benchmarks including APPS, BigCodeBench and HumanEval, Qwen3-8B moves from 0.503 to 0.689 average. In matched policy-value controls, ReST-GRPO + VM-MCTS reaches 0.642 on APPS-500 while GRPO + VM-MCTS reaches only 0.538, so the stage-one distributional shift survives value learning. End-to-end accounting puts ReST-GRPO at 1,752 GPU-hours against 2,080 for GRPO, hitting a 9% gain in 71 hours instead of 207. A value model trained only on code trajectories also transfers to MATH, Omni-MATH and GPQA-Diamond without target-domain tuning.

Reinforcement LearningGRPOMonte Carlo Tree SearchSining Zhoubian, Dan Zhang, Jie Tang·Aug 27, 2025

RelateAnything: Real-Time Open-Vocabulary Relation Prediction From Any Inputs

Open-vocabulary detection accepts any class list at inference, and promptable segmentation returns regions without class names: the taxonomy has left the model and become an input. Relation prediction has not. Scene-graph models are still trained and evaluated on the 50 or 56 predicates of one annotation style, their relation head conditioned on object labels and so tied to one detector. Three obstacles explain this, none primarily modelling: no relation corpus is both free-text and verified, a label-conditioned architecture cannot accept a vocabulary it was not trained on, and the standard metric rewards agreement with the training corpus, so a larger vocabulary scores as a regression. We present RelateAnything, a 53M-parameter model taking an image and regions from any source and returning scored relations over a predicate vocabulary supplied at inference as strings. Object labels are never an input, so the region source can change without retraining, and the vocabulary is a bank of text embeddings, not a learned classifier. It runs at 20 ms/frame. Training over 19,103 predicates requires positive-unlabeled supervision and a text encoder that separates antonyms, which contrastive encoders embed at cosine 0.95. To supply the supervision we build RA-4M, 474k images and 4.3M relations over 10,102 free-text predicates, generated against numbered box markers and geometrically verified. To measure it we build OV-SGG-Bench, six axes scored across datasets that the priors standard recall rewards cannot satisfy. On three cross-dataset benchmarks and a fourth zero-shot, RelateAnything has 2.3-3.5x the mean recall of the strongest open-vocabulary method of comparable scale, margins that survive a real detector, and leads a 3B-VLM scene-graph model on both metrics at under 2% of its parameters. In-domain measurement overstates transfer gains ~5x. Model, corpus and benchmark are public.

Scene Graph GenerationOpen-vocabularyRelation PredictionNeau, Maëlic·Sep 11, 2026