Persistent Robot World Models: Stabilizing Multi-Step Rollouts via Reinforcement Learning

• URL: http://arxiv.org/abs/2603.25685v1
• Date: Thu, 26 Mar 2026 17:36:08 GMT
• Title: Persistent Robot World Models: Stabilizing Multi-Step Rollouts via Reinforcement Learning
• Abstract summary: Action-conditioned robot world models generate future video frames of the manipulated scene given a robot action sequence.<n>These models are optimized for short-term prediction and break down when deployed autoregressively.<n>We introduce a reinforcement learning scheme that trains the world model on its own autoregressive rollouts.
• Score: 18.397872306430006
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Action-conditioned robot world models generate future video frames of the manipulated scene given a robot action sequence, offering a promising alternative for simulating tasks that are difficult to model with traditional physics engines. However, these models are optimized for short-term prediction and break down when deployed autoregressively: each predicted clip feeds back as context for the next, causing errors to compound and visual quality to rapidly degrade. We address this through the following contributions. First, we introduce a reinforcement learning (RL) post-training scheme that trains the world model on its own autoregressive rollouts rather than on ground-truth histories. We achieve this by adapting a recent contrastive RL objective for diffusion models to our setting and show that its convergence guarantees carry over exactly. Second, we design a training protocol that generates and compares multiple candidate variable-length futures from the same rollout state, reinforcing higher-fidelity predictions over lower-fidelity ones. Third, we develop efficient, multi-view visual fidelity rewards that combine complementary perceptual metrics across camera views and are aggregated at the clip level for dense, low-variance training signal. Fourth, we show that our approach establishes a new state-of-the-art for rollout fidelity on the DROID dataset, outperforming the strongest baseline on all metrics (e.g., LPIPS reduced by 14% on external cameras, SSIM improved by 9.1% on the wrist camera), winning 98% of paired comparisons, and achieving an 80% preference rate in a blind human study.

Related papers

• SC3-Eval: Evaluating Robot Foundation Models via Self-Consistent Video Generation [73.2142090645987]
  SC3-Eval is a self-consistent video generation recipe that adapts a pre-trained video foundation model into an accurate policy evaluator.<n>We demonstrate SC3-Eval rollouts reproduce the failure modes that policies exhibit in real-world rollouts, supporting fine-grained diagnostic comparison.
  arXiv  Detail & Related papers  (2026-06-17T02:15:46Z)
• RoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models [39.60337316193729]
  We present RoboAlign-R1, a framework that combines reward-aligned post-training with stabilized long-horizon inference for robot video world models.<n>Results show that reward-aligned post-training and stabilized long-horizon decoding improve task consistency, physical realism, and long-horizon prediction quality.
  arXiv  Detail & Related papers  (2026-05-05T14:49:00Z)
• Unified 4D World Action Modeling from Video Priors with Asynchronous Denoising [22.899605451385824]
  We propose X-WAM, a Unified 4D World Model that unifies real-time robotic action execution and high-fidelity 4D world synthesis (video + 3D reconstruction) in a single framework.<n>To leverage the strong visual priors of pretrained video diffusion models, X-WAM imagines the future world by predicting multi-view RGB-D videos.<n>Asynchronous Noise Sampling (ANS) to jointly optimize generation quality and action decoding efficiency.
  arXiv  Detail & Related papers  (2026-04-29T14:01:54Z)
• Disentangled Robot Learning via Separate Forward and Inverse Dynamics Pretraining [28.30092786035367]
  DeFI is a novel framework that Decouples visual Forward and Inverse dynamics pretraining to exploit respective data sources.<n>We introduce the General Forward Dynamics Model (GFDM), pretrained on diverse human and robot videos for future prediction, and the General Inverse Dynamics Model (GIDM), trained via self-supervised learning to infer latent actions from unlabeled video transitions.<n>Experiments on CALVIN ABC-D and SimplerEnv demonstrate state-of-the-art performance, with DeFI achieving an average task length of 4.51 for CALVIN, 51.2% success rate on SimplerEnv-Frac
  arXiv  Detail & Related papers  (2026-03-27T17:20:10Z)
• VLAW: Iterative Co-Improvement of Vision-Language-Action Policy and World Model [87.75549463328836]
  The goal of this paper is to improve the performance and reliability of vision-language-action (VLA) models through iterative online interaction.<n>We propose a simple iterative improvement algorithm that uses real-world roll-out data to improve the fidelity of the world model.<n>We achieve a 39.2% absolute success rate improvement over the base policy and 11.6% improvement from training with the generated synthetic rollouts.
  arXiv  Detail & Related papers  (2026-02-12T15:21:47Z)
• InternVLA-A1: Unifying Understanding, Generation and Action for Robotic Manipulation [77.07565723756119]
  We present InternVLA-A1, a vision-language-action model with dynamic predictive capabilities.<n>We pre-train these models on heterogeneous data sources over real-world robot data, synthetic simulation data, and human videos.<n>We evaluate InternVLA-A1 on 12 real-world robotic tasks and a simulation benchmark.
  arXiv  Detail & Related papers  (2026-01-05T18:54:29Z)
• Rethinking Training Dynamics in Scale-wise Autoregressive Generation [22.58390823803937]
  Next-scale prediction has emerged as a popular paradigm, where models generate images in a coarse-to-fine manner.<n>Scale-wise AR models suffer from exposure bias, which undermines generation quality.<n>We propose Self-Autoregressive Refinement to address these limitations.
  arXiv  Detail & Related papers  (2025-12-06T12:41:42Z)
• Reinforcement Learning with Inverse Rewards for World Model Post-training [29.19830208692156]
  We propose Reinforcement Learning with Inverse Rewards to improve action-following in video world models.<n>RLIR derives verifiable reward signals by recovering input actions from generated videos using an Inverse Dynamics Model.
  arXiv  Detail & Related papers  (2025-09-28T16:27:47Z)
• Physical Autoregressive Model for Robotic Manipulation without Action Pretraining [65.8971623698511]
  We build upon autoregressive video generation models to propose a Physical Autoregressive Model (PAR)<n>PAR leverages the world knowledge embedded in video pretraining to understand physical dynamics without requiring action pretraining.<n>Experiments on the ManiSkill benchmark show that PAR achieves a 100% success rate on the PushCube task.
  arXiv  Detail & Related papers  (2025-08-13T13:54:51Z)
• Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion [67.94300151774085]
  We introduce Self Forcing, a novel training paradigm for autoregressive video diffusion models.<n>It addresses the longstanding issue of exposure bias, where models trained on ground-truth context must generate sequences conditioned on their own imperfect outputs.
  arXiv  Detail & Related papers  (2025-06-09T17:59:55Z)
• VidMan: Exploiting Implicit Dynamics from Video Diffusion Model for Effective Robot Manipulation [79.00294932026266]
  VidMan is a novel framework that employs a two-stage training mechanism to enhance stability and improve data utilization efficiency. Our framework outperforms state-of-the-art baseline model GR-1 on the CALVIN benchmark, achieving a 11.7% relative improvement, and demonstrates over 9% precision gains on the OXE small-scale dataset.
  arXiv  Detail & Related papers  (2024-11-14T03:13:26Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.