Event
AWAM releases active-view imagination method and announces occlusion task results
1 reports · 1 sources
Jiawei Geng and others propose the Active World-Action Model (AWAM) and report success rates on occlusion benchmarks and real-robot trials: LIBERO-Occ 75.9%, LIBERO-Plus 62.4%, RLBench-OG Occlusion1 76.5%, and real-robot trials 94.4%. The paper states that these results support better robustness under the occlusion and distribution shifts evaluated. The method has the model actively plan observation positions and, before execution, plan trajectories and imagine what will be seen: predicted task-relevant virtual camera motion is used as a structured query to a geometry-conditioned world model, yielding a reusable spatial prior that is compressed into a fixed memory and fused with real-time observations via cross-attention for diffusion-based action prediction.
Report timeline
- AWAM: Active Trajectory-Conditioned View Imagination for Robust Robot Manipulation
AWAM (Active World-Action Model) introduces a virtual active-perception paradigm that actively plans observation positions and, before execution, imagines what will be seen along the planned trajectory. It treats the predicted task-related virtual camera motion as a structured query into a geometry-conditioned world model, yielding a reusable spatial prior that is compressed into a fixed memory and fused with live observations through cross-attention for diffusion-based action prediction.
arXiv cs.RO ·