Gravity Control: Real-Time Correction of Robotic Actions with Natural Language Intervention
一块面包，就让全球机器人集体破防？幕后真相终于被这家黑马破解
The article describes how the Gravity Control model, developed by Geek+, enables real-time correction of robotic actions using natural language prompts without retraining. This model identifies and manipulates key features within the robot's decision-making process to adjust actions on the fly, significantly improving success rates in out-of-distribution tasks from 17.5% to 87.5%, with specific tasks like horizontal bread placement achieving 95% success. The approach offers a practical solution to localized action deviation in deployed robots.
The article highlights a novel approach to real-time correction of robotic actions using the Gravity Control model, which allows for natural language intervention without retraining. This method significantly improves success rates in out-of-distribution tasks, offering a practical solution to the long-standing challenge of localized action deviation in deployed robots.
Source: AI Era — Robotics on WeChat · 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
5%
Reported success rate
17.5%
View original evidence
结果是：在真实的分布外复杂操作任务测试中，整体成功率从 17.5% 大幅提升至 87.5%；其中横放面包这一单项 OOD 设置，直接由 5% 飙升至 95%！
Open source S11Reported success rate
87.5%
View original evidence
结果是：在真实的分布外复杂操作任务测试中，整体成功率从 17.5% 大幅提升至 87.5%；其中横放面包这一单项 OOD 设置，直接由 5% 飙升至 95%！
Open source S11Reported success rate
56%
Reported zero-shot result; the excerpt does not specify an evaluation protocol.
View original evidence
Figure AI 在 2026 年初推出 Helix 02，随后展示了机器人在完全自主控制下连续跑满 8 小时班次的能力；其 Index 预训练将零样本成功率从约 9% 提升至 56%。
Open source S21Reported success rate
≈9%
Reported zero-shot result; the excerpt does not specify an evaluation protocol.
View original evidence
Figure AI 在 2026 年初推出 Helix 02，随后展示了机器人在完全自主控制下连续跑满 8 小时班次的能力；其 Index 预训练将零样本成功率从约 9% 提升至 56%。
Open source S21
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: 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.
在这个「横放面包」的测试设置里，原始策略模型的成功率只有 5%。
Open source S4
结果是：在真实的分布外复杂操作任务测试中，整体成功率从 17.5% 大幅提升至 87.5%；其中横放面包这一单项 OOD 设置，直接由 5% 飙升至 95%！
Open source S11
Figure AI 在 2026 年初推出 Helix 02，随后展示了机器人在完全自主控制下连续跑满 8 小时班次的能力；其 Index 预训练将零样本成功率从约 9% 提升至 56%。
Open source S21
Source：AI Era — Robotics on WeChat · mp.weixin.qq.com