Research
115d agoGRASP Enables Long-Horizon World Model Planning
The GRASP planner optimizes trajectories by lifting them into virtual states, allowing parallel optimization across time. This approach solves the vanishing gradient problem common in long-horizon world models. Researchers at BAIR added stochasticity to the process to improve robustness.
The Signal
Practitioners can now execute complex, multi-step plans without the typical computational collapse.