RISE: Self-Improving Robot Policy with Compositional World Model

2026-02-11

AI Summary

Existing vision-language-action models often fail in dynamic, contact-intensive robotic manipulation due to minor deviations, while real-world online reinforcement learning is hindered by safety concerns, high costs, and environment reset challenges. To address these limitations, this work proposes RISE—a framework that enables closed-loop self-improvement entirely in imagination through a composable world model. RISE leverages a controllable multi-view dynamics model to predict future states and employs a disentangled progress-value model to evaluate outcomes and generate advantage signals, thereby driving policy gradient updates without any physical interaction. Evaluated on three real-world tasks—dynamic block sorting, backpack packing, and box sealing—the method substantially outperforms existing approaches, achieving performance gains of 35%, 45%, and 35%, respectively.

Technology Category

Intelligent Robots: ManipulationMachine Learning: Imitation Learning & Inverse Reinforcement LearningHumans and AI: Human-Aware Planning and Behavior Prediction

Application Category

Responsible Web: Machine-in-the-loop, human agency and autonomySearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Despite the sustained scaling on model capacity and data acquisition, Vision-Language-Action (VLA) models remain brittle in contact-rich and dynamic manipulation tasks, where minor execution deviations can compound into failures. While reinforcement learning (RL) offers a principled path to robustness, on-policy RL in the physical world is constrained by safety risk, hardware cost, and environment reset. To bridge this gap, we present RISE, a scalable framework of robotic reinforcement learning via imagination. At its core is a Compositional World Model that (i) predicts multi-view future via a controllable dynamics model, and (ii) evaluates imagined outcomes with a progress value model, producing informative advantages for the policy improvement. Such compositional design allows state and value to be tailored by best-suited yet distinct architectures and objectives. These components are integrated into a closed-loop self-improving pipeline that continuously generates imaginary rollouts, estimates advantages, and updates the policy in imaginary space without costly physical interaction. Across three challenging real-world tasks, RISE yields significant improvement over prior art, with more than +35% absolute performance increase in dynamic brick sorting, +45% for backpack packing, and +35% for box closing, respectively.

Problem

• Vision-Language-Action models
• contact-rich manipulation
• dynamic manipulation tasks
• on-policy reinforcement learning
• physical world constraints

Innovation

• Compositional World Model
• Imaginative Reinforcement Learning
• Vision-Language-Action Models
• Self-Improving Policy
• Model-Based Robotics