ForceHarness released
The story so far
Hugging Face has released ForceHarness, enabling frozen models to achieve action and physical interaction through force without demonstrations or fine-tuning. The tool, using wrist sensors or joint torque calibration, boosts the zero-shot success rate of GPT-6 Astra on real robotic arms from 3.3% to 90.0%, outperforms force-perception strategies based on demonstration training in the HiddenPhys benchmark, and matches handcrafted expert controllers. ForceHarness addresses the decision delay and force understanding issues of frozen models in contact interactions through self-terminating force primitives and a physics translation layer.
AI-written from the reports · Updated 16 hours ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hugging Face Daily PapersForceHarness: Turning Force into Action and Physics for Frozen Frontier Models
ForceHarness enables frozen models to leverage force without demonstration or fine-tuning, after calibration with arm sensors or joint torques, it boosts the zero-shot success rate of GPT-6 Astra from 3.3% to 90.0% on real robotic arms, surpasses force-perception strategies based on demonstration training in the HiddenPhys benchmark, and matches handcrafted expert controllers.
Heat over time
- Now
- 6
- Peak
- 818 Nov, 14:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.