GRASP: Gradient-Based Planning for World Models

Berkeley researchers introduce GRASP, a gradient-based planner for learned world models that extends planning horizons, improving long-term decision-making in AI systems.

One continuously updated timeline instead of dozens of separate articles. New developments are appended as the story evolves.

1. AnnouncementApr 20, 2026, 09:00 AM 84%

   Berkeley researchers unveil GRASP, a gradient-based planner for learned world models enabling longer-horizon planning.

   Berkeley researchers introduce GRASP, a gradient-based planner for learned world models that extends planning horizons, improving long-term decision-making in AI systems.

   Read the full story →