Event
RTPolicy Multi-View Visual Motion Policy Paper Released
1 reports · 1 sources
Xianbao Hou and other authors released a paper proposing RTPolicy, a multi-view visual motion policy that aggregates multi-camera information into a compact token representation through view-aware feature fusion and visual compression, and uses a flow-matching action generator to reduce action generation overhead. The authors claim that on 50 tasks in the RoboTwin 2.0 single-task clean evaluation, RTPolicy achieves an average success rate of 52.26%, higher than the strongest baseline on the leaderboard, π0, at 46.4%. In asynchronous streaming inference evaluation on a real dual-arm platform, the authors claim an average success rate of 72.0%, achieving real-time inference at 30.35 milliseconds (32.9 Hz) with INT16 precision on RDK S600, and 149.72 milliseconds (6.7 Hz) on RDK S100.
Report timeline
- RTPolicy: Real-time multi-view visuomotor policy for edge devices, inference at 32.9 Hz
RTPolicy is a multi-view visuomotor policy that uses view-aware feature fusion and visual compression to aggregate multi-camera information into a compact token representation, and uses a flow-matching action generator to reduce action-generation overhead. On the 50 tasks of the RoboTwin 2.0 single-task clean evaluation, it achieved an average success rate of 52.26%, higher than the strongest leaderboard baseline π0 at 46.4%.
arXiv cs.RO ·
Related topics
Interest trend
- Current
- 3
- Peak
- 4Nov 4 01:00
- Past 24 hours
- –
The trend compares participants observed throughout the same period, so its range may be smaller than the current interest count. Move or tap to inspect each hour; use the arrow keys on a keyboard.