Gravity Control: Real-Time Correction of Robotic Actions with Natural Language I
Overview
Geek+'s Gravity Control model enables real-time robotic action correction via natural language prompts, boosting success rates from 17.5% to 87.5% in out-of-distribution tasks, with one task achieving 95% success.
Generated from attributed reports · Updated 40 minutes ago
Event evidence and corrections
1 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · Reported success rate: 5 percent · Basis not reported · Differing source assertions
“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%。”
Exact source · revision 1aiera
Reported quantity · Reported success rate: 17.5 percent · Basis not reported · Differing source assertions
“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%！”
Exact source · revision 1aiera
Reported quantity · Reported success rate: 87.5 percent · Basis not reported · Differing source assertions
“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%！”
Exact source · revision 1aiera
Reported quantity · Reported success rate: 56 percent · Basis not reported · Reported zero-shot result; the excerpt does not specify an evaluation protocol. · Differing source assertions
“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%。”
Exact source · revision 1aiera
Reported quantity · Reported success rate: 9 percent · Basis not reported · Reported zero-shot result; the excerpt does not specify an evaluation protocol. · Differing source assertions
“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%。”
Exact source · revision 1aiera
Evidence dependency changes (1)
- 2026-10-30T17:19:47.614Z · evidence updated · source revision 1. Evidence extraction was updated; current source attributions are shown above.
Report timeline
Follow attributed reports and material updates.
- AI Era — Robotics on WeChatSignalGravity Control: Real-Time Correction of Robotic Actions with Natural Language Intervention
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.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).