D-Robotics and the University of Tokyo Release RawVLA Approach
What happened
D-Robotics, together with the University of Tokyo and other institutions, released the RawVLA technical approach: a learnable neural ISP module is added before the VLA policy, the VLA model is frozen throughout training, and only the action error is used to backpropagate to train the imaging front end. On real robots, π0.5 was used to test four dual-arm tasks. Under low light, the average success rate rose from 0% with the default ISP to 66.5%, including 92% for pick-and-place and 88% for drawer operation, while the DarkISP approach reached 11%; under normal lighting, RawVLA achieved 75.0% and the default ISP 76.5%, close to each other. The paper, code, and dataset have all been made publicly available.
AI 根据报道生成 · 6 days ago更新
Reporting timeline
Follow the reporting to explore different perspectives.
- 未来机器人网 · 微信公众号FeaturedD-Robotics and University of Tokyo release RawVLA: frozen VLA, only imaging front end trained, low-light dual-arm task success rate rises from 0 to 66.5%
D-Robotics, together with the University of Tokyo and other institutions, released the RawVLA technical approach, which adds a learnable neural ISP module before the VLA policy while keeping the VLA model frozen, using action error to train the imaging front end in reverse. On real robots, four dual-arm tasks were tested with π0.5; the average success rate in low light rose from 0% with the default ISP to 66.5%, versus 75.0% in normal light, where the default ISP achieved 76.5%.
Activity for this event
- Current
- 3
- Peak
- 8Nov 5 01:00
- Past 24 hours
- -50%
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.