Spotter: VLM intervenes only on failure, boosting real-robot success from 53% to 83%
青稞AI · wechat · 2026-10-11
Researchers from Griffith University, Tsinghua and partners present Spotter, addressing a core VLA weakness: embodied models keep executing even after failures and can't self-correct (under 1/10 of failures fixed even with hints; scorers perform near-randomly on failure states).
- Spotter flips the popular VLM-Led design: the VLA executes continuously, a local Qwen model screens in parallel for risk, and a VLM (Qwen or GPT) steps in only after two consecutive alerts to write a short repair plan, then reflect and hand back control
- RoboCasa: CosmosPolicy 67.9→73.5, π0.5 64.3→71.8 (GPT); RoboTwin2.0 Hard π0.5 up to 57.0 (+9.8)
- Successful episodes take only 10s longer; 70% faster than VLM-Led
- Real-robot Franka experiments: 53%→83% average success; long-horizon task 30%→70%
Runs locally with a single Qwen model (FP8); code open-sourced. (paper)
More from Embodied
- MIT CSAIL's X-Hinges 3D-Prints Everyday Objects With Built-In Sensing Across Three Motion Axes — MIT News AI · 2026-10-23
- Formic raises $50M Series B to scale physical AI across 100+ US factories — garrytan · 2026-10-23
- fal starts fine-tuning a physical robot to draw, results next week — OdinLovis · 2026-10-23
- PEARL demo: an LLM agent that checks experiment results, revises plans, and re-tests — KevinKaichuang · 2026-10-23
- CU Boulder's ATLAS Institute Opens Tenure-Track Robotics Professorship — ryosuzk · 2026-10-23
- MiTaS multi-resolution tactile imitation learning hits 80% success vs 35.5% vision-only baseline — Jan_R_Peters · 2026-10-23