arXiv · ML· Greta Perez-Haiek (New York University), Luca Macesanu (New York University), Kevin Angers (New York University), Mohammed Imam (New York University), Timothy Gambin (University of Malta), Christopher Clark (New York University)·· 6 小时前AI 评分23
Embodied at Sea: Imitation Learning for Multi-Task Underwater Manipulation
Embodied at Sea: Imitation Learning for Multi-Task Underwater Manipulation
AI 导读
summary_zh: 研究收集1616集水下操作数据训练14个VLA模型，在泳池任务中最佳模型成功率为93.3%。 案例研究显示单任务ACT策略（47.9%）显著优于通用SmolVLA（7.9%），相关工具已开源。
来源：arXiv · ML · arxiv.org