SenseTime's SenseNova-RoboRSI Doubles Robot Task Scores Without Retraining Models
JaynitMakwana · x · 2026-10-10
SenseTime introduced SenseNova-RoboRSI, a recursive self-improvement framework for embodied agents that upgrades the agent harness instead of retraining foundation models.
- RoboDojo average score of 56.83 with a 50.83% success rate, a 96.2% relative increase over the published baseline of 28.97, with model weights unchanged
- RSI loop: execute → diagnose → improve → validate → adopt, evolving harnesses from physical task failure feedback
- Two key techniques: multi-agent collaboration (Planning, Main, and Feedback subagents) and multi-point EEF prediction generating short action sequences for motion continuity
- The framework will be open-sourced with a full technical report to follow
More from Embodied
- NVIDIA 推出 Jetson Agent Skills，让编码 Agent 懂边缘硬件 — blaizedsouza · 2026-10-11
- Open-source New York City built in UE5 and Blender, with three.js rendering, ready for games and RL — DevDminGod · 2026-10-11
- Witnessing a scooter accident up close, one driver argues FSD matters because human attention fails — chrisfleck · 2026-10-11
- Multitap opens early access: Quest 3 app rebuilds classic games into 3D mixed reality live — gwendall · 2026-10-10
- China unveils AI smart blackboard that turns a cube into an interactive 3D lesson live — iamfakhrealam · 2026-10-10
- Autonomous massage robot needs zero hands — and it's 'EXTREMELY aggressive' — abhishekcode42 · 2026-10-10