Event
CaPtain: Small VLMs Post-Trained into Code-as-Policy Robot Agents
1 reports · 1 sources
Yijia Fan and co-authors released CaPtain, an open post-training pipeline that trains small open vision-language models into code-as-policy (CaP) robot agents that do not rely on VLAs, frontier models, or privileged state at test time, and they released all code, data, and model weights. The pipeline first uses synthetic trajectories generated by GPT-6 Astra for supervised fine-tuning, then performs multi-round reinforcement learning, and introduces a physics critic that reads simulator privileged state to assign credit to each request. The authors report that starting from Qwen3.5-4B (0% zero-shot success rate), CaPtain-4B reached 55.4% success on held-out LIBERO tasks after SFT and rose to 73.8% after reinforcement learning; after pure simulation adaptation it can run on a real robot.
Report timeline
- CaPtain: A pipeline for post-training small open VLMs into code-as-policy robot agents
CaPtain proposes an open post-training pipeline that trains small open vision-language models (VLMs) into code-as-policy (CaP) robot agents, without relying on VLAs, frontier models, or privileged state at test time. The pipeline first performs supervised fine-tuning on synthetic trajectories generated by GPT-6 Astra, then applies multi-turn reinforcement learning, introducing a physics critic that reads simulator privileged state to assign credit to each request.
arXiv cs.RO ·
Related topics
Interest trend
- Current
- 3
- Peak
- 4Nov 4 01:00
- Past 24 hours
- –
The trend compares participants observed throughout the same period, so its range may be smaller than the current interest count. Move or tap to inspect each hour; use the arrow keys on a keyboard.