ReSteer open-sourced: fixing VLA policies that ignore mid-execution instruction switches
siddkaramcheti · x · 2026-10-06
A Georgia Tech team's RSS 2026 paper ReSteer shows frontier VLA robot policies are far less steerable than advertised: switch the instruction mid-execution and models like π0.5 plow ahead with the original task. The paper quantifies steerability using conditional mutual information I(A;L|s) to find states where policies ignore language, then generates data that fixes the failure mode. The full LIBERO pipeline is now open-sourced on GitHub.
More from Embodied
- The robotics data business's biggest bottleneck is operational, not technical — paigeinsf · 2026-10-06
- GPT-2-powered robot Samson explores, gets a little confused but has fun — MikePFrank · 2026-10-06
- VC: almost none of real robot deployment data flows back into training — carrycooldude · 2026-10-06
- RealtimeWAM: One-Step Asynchronous World Action Model Delivers 25x Speedup with <1% Accuracy Loss — NanyangTechnologicalUniversity · 2026-10-06
- ACG-Bench Probes Dual-Arm VLA Generalization; AE-VLA Lifts Success from 3% to 21.5% — Zaibin Zhang · 2026-10-06
- DIY robotic backboard uses computer vision to make every beginner shot go in — TinfoilTricorn · 2026-10-06