Yann LeCun's retweeted world model planner GRASP code open-sourced

MeNews · 2026-04-05T07:32:25+00:00

On April 1st, the GRASP code was open-sourced. It is a parallel stochastic gradient planner used for learning world models, optimizing actions and intermediate states. GRASP achieves efficient planning through dynamic consistency penalties, features multiple key designs, is suitable for various environments, and can be integrated with other models.

MeNews

2026-04-05 07:32:25

Abstract generation in progress

ME News update: On April 1 (UTC+8), the GRASP code for the new world model planner published by Michael Psenka and reposted by Yann LeCun has been open-sourced. GRASP is a parallel stochastic gradient planner for learning world models. Its core idea is to simultaneously optimize actions and intermediate “virtual” states in an improved trajectory space. It uses a dynamics consistency penalty to ensure consistency with the learned dynamics model, and enables efficient parallelized planning. Key design elements include introducing virtual states that can be computed in parallel, exploring the state space via noise, using stop-gradient techniques to prevent gradient flow toward high-dimensional visual model state inputs, and combining target shaping with periodic serial expansion synchronized steps to ensure stability. The planner is built on the DINO-WM project, but it can also be integrated with other world models (such as LeWorldModel, JEPA-WMs) through adapters, making it applicable to a variety of environments like PushT and TwoRoom. The codebase provides the core algorithm implementations and adapter installation scripts. (Source: InFoQ)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.