Event
Metro-WM replaces latent subgoals with real-state retrieval
1 reports · 1 sources
Royson Lee and colleagues propose Metro-WM, a hierarchical planning framework that, instead of generating unconstrained latent subgoals, retrieves real states from observed frames in offline expert demonstrations or random action trajectories as subgoals, and connects frames from different episodes into a path toward the goal. The authors say the starting point for this design is that existing leading macro-planners frequently output physically unrealizable subgoals. Experiments show that Metro-WM's long-horizon success rate is up to 37.33 percentage points higher than the next-best hierarchical method, up to 10.9 times faster, with 13–56 times less offline computation, and with less tuning; when the micro-planner deviates from the route, it can immediately find a new optimal path from the current state. The authors also claim the method remains robust under extremely sparse data conditions.
Report timeline
- Metro-WM: Long-horizon latent planning with realizable sub-goals
Metro-WM proposes a hierarchical planning framework that, instead of generating unconstrained latent sub-goals, retrieves real states from observed frames in offline expert demonstrations or random-action trajectories as sub-goals, and connects frames from different episodes into paths toward the goal. Experiments show its long-horizon success rate is up to 37.33 percentage points higher than the next-best hierarchical method, up to 10.9 times faster, uses 13-56 times less offline computation, and requires less tuning. The authors also note that leading macro planners frequently output physically unrealizable sub-goals.
arXiv cs.RO ·