Event
Geek+ Releases Gravity Control Causal Intervention Model
2 reports · 2 sources
Under its Gravity embodied intelligence framework, Geek+ has released the Gravity Control causal intervention model, which can intervene on the model's internal latent features at inference time via natural language, without updating the original policy weights or adding new demonstration data, to correct robot actions at runtime. Geek+ claims that the model raises the success rate on OOD tasks from 5% to 95%. The model has been evaluated on VLA architectures such as π0.5 and OpenVLA-OFT, on the LIBERO simulation benchmark, and on physical robot platforms. Geek+ states that its correction presupposes that the base policy itself already possesses basic task capability, and that the goal is not simply to push actions in a certain direction; the system also introduces closed-loop gating that sets an upper limit on the degree of intervention.
Report timeline
- Geek+ releases Gravity Control causal intervention model to correct robot actions online with natural language
Under its Gravity embodied intelligence framework, Geek+ has launched the Gravity Control causal intervention model, which can correct local action deviations in robots with a single natural-language prompt, without updating the original policy weights or adding new demonstration data.
新智元 · 微信公众号 ·
- Geek+ Gravity Control uses natural language for runtime correction, raising OOD task success rate from 5% to 95%
Under the Gravity framework, Geek+ embodied intelligence proposes the Gravity Control causal intervention model. Without training the original policy model or adding new demonstration data, it uses mechanistic interpretability to uncover the model's internal action regulation mechanisms and intervenes on relevant latent features at inference time to achieve runtime correction.
OFweek Robotics ·