Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2503.09871 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909536242106368 |
|---|---|
| author | Qiu, Xiaowen Wang, Yian Cai, Jiting Chen, Zhehuan Lin, Chunru Wang, Tsun-Hsuan Gan, Chuang |
| author_facet | Qiu, Xiaowen Wang, Yian Cai, Jiting Chen, Zhehuan Lin, Chunru Wang, Tsun-Hsuan Gan, Chuang |
| contents | Automatically generating training supervision for embodied tasks is crucial, as manual designing is tedious and not scalable. While prior works use large language models (LLMs) or vision-language models (VLMs) to generate rewards, these approaches are largely limited to simple tasks with well-defined rewards, such as pick-and-place. This limitation arises because LLMs struggle to interpret complex scenes compressed into text or code due to their restricted input modality, while VLM-based rewards, though better at visual perception, remain limited by their less expressive output modality. To address these challenges, we leverage the imagination capability of general-purpose video generation models. Given an initial simulation frame and a textual task description, the video generation model produces a video demonstrating task completion with correct semantics. We then extract rich supervisory signals from the generated video, including 6D object pose sequences, 2D segmentations, and estimated depth, to facilitate task learning in simulation. Our approach significantly improves supervision quality for complex embodied tasks, enabling large-scale training in simulators. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_09871 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LuciBot: Automated Robot Policy Learning from Generated Videos Qiu, Xiaowen Wang, Yian Cai, Jiting Chen, Zhehuan Lin, Chunru Wang, Tsun-Hsuan Gan, Chuang Computer Vision and Pattern Recognition Automatically generating training supervision for embodied tasks is crucial, as manual designing is tedious and not scalable. While prior works use large language models (LLMs) or vision-language models (VLMs) to generate rewards, these approaches are largely limited to simple tasks with well-defined rewards, such as pick-and-place. This limitation arises because LLMs struggle to interpret complex scenes compressed into text or code due to their restricted input modality, while VLM-based rewards, though better at visual perception, remain limited by their less expressive output modality. To address these challenges, we leverage the imagination capability of general-purpose video generation models. Given an initial simulation frame and a textual task description, the video generation model produces a video demonstrating task completion with correct semantics. We then extract rich supervisory signals from the generated video, including 6D object pose sequences, 2D segmentations, and estimated depth, to facilitate task learning in simulation. Our approach significantly improves supervision quality for complex embodied tasks, enabling large-scale training in simulators. |
| title | LuciBot: Automated Robot Policy Learning from Generated Videos |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.09871 |