Event
UC Davis and ADI Open-Source Agile-WAM Tactile Manipulation Model
1 reports · 1 sources
The UC Davis and ADI team proposed and open-sourced Agile-WAM, a tactile world-action model for contact-rich manipulation that couples multimodal world prediction with action generation. On five tasks on a real Flexiv Rizon 4 robot—gear assembly, peg insertion, network cable insertion/removal, and power plug insertion—overall success rate improved by 29.4% relative; on a single RTX 4090, inference latency is 10.35 ms and control frequency is 96.62 Hz, reportedly about 40 times higher than the diffusion policy DP. In the real-robot experiments, 50 expert demonstrations were collected per task via Meta Quest 3 teleoperation, and evaluation used only 20 trials per task; the paper itself states that statistical power is limited. Inference measurements were FP32, batch size 1. On 9 ManiFeel simulation tasks, the model achieved best or near-best results on multiple tasks.
Report timeline
- UC Davis and ADI open-source Agile-WAM tactile world action model: 29.4% relative improvement in real-robot success rate
UC Davis and ADI have open-sourced Agile-WAM, a tactile world action model for contact-rich manipulation. Through multi-scale, multimodal prediction, it lets vision be supervised over long spans and touch over next-frame spans, achieving a 29.4% relative improvement in overall success rate across five real-robot tasks. On a single RTX 4090, inference latency is 10.35 ms with a control frequency of 96.62 Hz, about 40 times higher than the diffusion policy DP; it achieves best or near-best results on multiple of 9 ManiFeel simulation tasks and Flexiv Rizon 4 real-robot tasks.
3D视觉工坊 · 微信公众号 ·