Event
Spotter: Embodied models execute autonomously, VLM corrects errors on demand
1 reports · 1 sources
Li Long, a PhD student at Griffith University, Zhao Qichao, a PhD student at Tsinghua University, together with teams from Fudan University and Tongji University, proposed the Spotter method: the embodied model continues to execute autonomously, a lightweight local model (always Qwen) supervises in parallel, and the VLM intervenes to correct errors only when they occur, so as to avoid the robot waiting for VLM inference at every step; it was evaluated in simulation and on real robots. In the evaluation, the step limit was uniformly set to 1.8 times the official task evaluation step count (900 to 1500 steps for the vast majority of tasks), and no privileged information was used in any case; Spotter's repair is premised on "the scene still being recoverable after an error occurs," and Qwen's time is calculated based on inference speed on a single H200 with only one request processed at a time.
Report timeline
- Spotter paper: Embodied model executes autonomously, VLM corrects errors on demand to reduce robot operation latency
Li Long, a PhD student at Griffith University, Zhao Qichao, a PhD student at Tsinghua University's Institute for Interdisciplinary Information Sciences, and teams from Fudan University and Tongji University propose Spotter: an embodied model executes continuously and autonomously while a lightweight local model supervises in parallel, and the VLM intervenes to correct errors only when they occur, avoiding the robot waiting for VLM inference at every step.
Xbotics具身智能实验室 · 微信公众号 ·