Research
JEPA world models + Hierarchical Planning is a massive step for long-horizon robotics. A classic failure mode I’ve faced with planning with …
JEPA world models + Hierarchical Planning is a massive step for long-horizon robotics. A classic failure mode I’ve faced with planning with world models: flat planning often 'cheats.' For example, in
JEPA world models + Hierarchical Planning is a massive step for long-horizon robotics. A classic failure mode I’ve faced with planning with world models: flat planning often "cheats." For example, in pick-and-place, the robot often reaches the target state in imagination without actually picking the object. Hierarchical layers fix this by first optimizing for valid subgoals (like the grip) before the finish. Incredible work, @kevinghstz and team! Huge congratulations. 🚀 Hierarchical planning unlocks long-horizon, non-greedy behavior in JEPA world models. Paper: https://arxiv.org/pdf/2604.03208 Website: https://kevinghst.github.io/HWM/ Code: https://github.com/kevinghst/HWM_PLDM
Related
- Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models
- Reasoning Fails Where Step Flow Breaks
- On the Step Length Confounding in LLM Reasoning Data Selection
- Optimal Decay Spectra for Linear Recurrences
- AdaSpark: Adaptive Sparsity for Efficient Long-Video Understanding
Source: Yann LeCun (X) | 2026-04-08