作者位次 #3/7ACL 2026Poster· Natural Language Processing
On-policy Reinforcement Fine-tuning with Offline reward for Multi-step Embodied Planning
Di Wu · Jiaxin Fan · Chloe Gu · Guanbo Wang · Wei Yin · Wenhao Li · Bo Jin
摘要
Embodied planning requires agents to make coherent multi-step decisions based on dynamic visual observations and verbal goals. While recent vision-language models (VLMs) excel at static perception tasks, they struggle in interactive environments. Reinforcement learning (RL) offers a natural way t...