OpenDriveLab/RISE

[RSS 2026] Code for RISE: Self-Improving Robot Policy with Compositional World Model

What it solves

RISE addresses the high cost and laborious reset processes associated with real-world robot interactions. It provides a framework for robot policies to improve themselves through simulated experiences (imagination) rather than relying solely on physical hardware interactions.

How it works

RISE uses a compositional world model consisting of a controllable multi-view dynamics model and a progress value model. This setup allows the system to generate imaginary rollouts—simulated sequences of actions and outcomes—which are then used to train and bootstrap robot policies via online reinforcement learning.

Who it’s for

Researchers and developers working on robotic manipulation and reinforcement learning who want to scale policy improvement without the need for constant physical resets or expensive hardware wear.

Highlights

  • Compositional World Model: Combines dynamics and progress value models to provide informative advantages for policy training.
  • RL in Imagination: Enables scalable self-improvement by training policies in a simulated environment created by the world model.
  • Real-world Gains: Demonstrated significant performance increases in dexterous tasks like brick sorting (+35%), backpack packing (+45%), and box closing (+35%).

Related

  • Project
    fan-ziqi/rl_sar

    A framework for simulation verification and physical deployment of reinforcement learning policies for quadruped, wheeled, and humanoid robots.

  • Project
    isaac-sim/IsaacLab-Arena

    Isaac Lab‑Arena is an alpha‑stage, open‑source extension to NVIDIA Isaac Lab that lets you build robot simulation environments by mixing reusable *scene*, *embodiment*, and *task* primitives. It assembles these at runtime, supports sequential task chaining, natural‑language object placement, and large‑scale parallel evaluation, making it easier to benchmark or train generalist robot policies. Installation works via the `uv` Python manager or a provided Docker image; the library is Linux‑only and requires an NVIDIA GPU and Isaac Sim. The project is under Apache 2.0 (with proprietary Isaac Sim components) and is intended for research rather than production.

  • Project
    ReinFlow/ReinFlow

    A policy gradient framework for fine-tuning flow matching policies using online reinforcement learning, enabling efficient optimization of robotic control policies with minimal denoising steps.

  • Project
    FrontisAI/OpenRSI

    OpenRSI is an open initiative and framework for recursive self-improvement, providing the OpenMLE stack and Frontis-MA1 model to enable AI to autonomously improve its own machine learning engineering capabilities.

  • Project
    RoboDojo-Benchmark/RoboDojo

    A unified sim-and-real benchmark for evaluating generalist robot manipulation policies across 60 tasks and five capability dimensions.