AnalysisRoboticsOctober 16, 2026

ORA harness lets open-weight models operate robots zero-shot

Read original source →haoailab.com

UCSD's Hao AI Lab reports ORA hits 69.2% on RoboTwin C2R (250 cases) and 81.4% on LIBERO-Pro (800 cases) with no VLA, robot training, or task-specific memory. It runs open-weight models like DeepSeek-V4.1-Flash, competitive with agents built on frontier models.

1 source

More stories today

Open the live feed