LIFT: 20-30 Force Traces Teach Pretrained VLA Models Reactive Force Control
机器之心 · wechat · 2026-09-27
A Shanghai Jiao Tong University team (Lu Cewu, Wen Chuan) with QunChu proposed LIFT (Late Reactive Injection of Force for VLA Post-Training), accepted at CoRL 2026 with high scores and open-sourced.
LIFT teaches pretrained VLA models force perception purely via online post-training—no force-data pretraining needed. On towel folding, book insertion, and Hananoi tasks it significantly outperforms vision-only post-training (e.g., 26.7→56.7) with only 20-30 online force traces. It preserves pretrained knowledge via action-expert weight copying and shifted causal attention masks, raises control-loop frequency from 1Hz to 10Hz, and mixes in force-free data to avoid overfitting. The authors argue the paradigm applies to any MoT-based pretrained model.
More from Embodied
- 5.46M-param touch model LingBot-Touch pokes straws through sealed lids autonomously — Puzzleheaded-Buy3008 · 2026-11-14
- OpenAI disbands robotics team five months after relaunch, leaves empty CoRL booth — ns123abc · 2026-11-14
- Fable 5.2 drives a robot arm to paint its self-portrait: 12 hours, 188M tokens, $97 — VoicAndrei · 2026-11-14
- 'Are robots just chatbots with arms?' Reddit debates what's really inside robot stacks — ironAtlasuy7 · 2026-11-14
- Wayve explains building safety as an intrinsic property of its AI driver — alexgkendall · 2026-11-14
- YC-backed Vexo unveils always-listening AI bracelet that turns conversations into actions — ycombinator · 2026-11-14