VISION-LANGUAGE MODELS AS TRAINERS FOR INSTRUCTION-FOLLOWING AGENTS
Keywords: Language-conditioned Reinforcement Learning, Reward Generation, Vision Language Foundation Models
Abstract: Developing agents that can understand and follow language instructions is critical for effective and reliable human-AI collaboration. Recent approaches train these agents using reinforcement learning with infrequent environment rewards, placing a significant burden on environment designers to create language-conditioned reward functions. As environments and instructions grow in complexity, crafting such reward functions becomes increasingly impractical. To address this challenge, we introduce V-TIFA, a novel method that trains instruction-following agents by leveraging feedback from vision-language models (VLMs). The core idea of V-TIFA is to query VLMs to rate entire trajectories based on language instructions, using the resulting ratings to directly train the agent. Unlike prior VLM reward generation methods, V-TIFA does not require manually crafted task specifications, enabling agents to learn from a diverse set of natural language instructions. Extensive experiments in embodied environments demonstrate that V-TIFA outperforms existing reward generation methods under the same conditions.
Primary Area: reinforcement learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7353
Loading