Ke Sun, Kevin Yuchen Ma, Yike Zhang, Yanzhe Chen, Mike Zheng Shou (Show Lab) · Force-aware robot harness

ForceHarness

ForceHarness is a harness that lets frozen, API-only multimodal models use contact force for robot manipulation without demonstrations or finetuning. It uses self-terminating force primitives that regulate force beneath the model, and a physics translation layer that turns force traces into physical quantities with uncertainty. The authors also introduce HiddenPhys, a seven-task MuJoCo benchmark with hidden physical variables. On a real arm without a wrist force sensor, it raises the zero-shot success rate of GPT-6 Astra from 3.3% to 90.0%, and adds 17.5 and 34.3 percentage points on ForceBench and HiddenPhys respectively.

Open the source ↗

Similar Agents & Automation releases

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.