Geek+ Unveils Gravity Control to Enable Natural Language Correction of Robot Actions During Inference
从5%到95%，极智嘉具身智能实现机器人“一句话纠偏”
Geek+ has introduced Gravity Control, a method that intervenes in the internal representations of robot models during inference using natural language prompts, without updating model weights or collecting new demonstration data. In tests on real robots with out-of-distribution tasks, the overall success rate improved from 17.5% to 87.5%, including a specific task involving horizontally placed bread that rose from 5% to 95%. The approach aims to address localized action deviations in deployed systems by allowing operators to adjust robot behavior through simple instructions like 'grasp deeper,' rather than requiring full retraining.
Source: Robotics — Chinese web discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
17.5%
Reported trials
View original evidence
在其测试的真实机器人 OOD 任务中，整体成功率从 17.5% 提升至 87.5%；其中，横放面包这一单项设置从 5% 提升至 95%。
Open source S11Reported success rate
87.5%
Reported trials
View original evidence
在其测试的真实机器人 OOD 任务中，整体成功率从 17.5% 提升至 87.5%；其中，横放面包这一单项设置从 5% 提升至 95%。
Open source S11
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
在其测试的真实机器人 OOD 任务中，整体成功率从 17.5% 提升至 87.5%；其中，横放面包这一单项设置从 5% 提升至 95%。
Open source S11
从5%到95%，极智嘉具身智能实现机器人“一句话纠偏”
Open source S1
Source：Robotics — Chinese web discovery · msn.cn