论文HF Daily Papers · arXiv 2611.189199 天前
VR人机协同修正点云抓取策略,硬样本成功率从24%升至45%
本文由 AI 根据公开内容整理,可能有误,请以原文为准:Correcting Persistent Failures in Point-Cloud Grasping Policies using Human Demonstrations in Virtual Reality(HF Daily Papers · arXiv 2611.18919)
核心内容
针对无模型点云抓取中强化学习策略存在的特定场景持续失效问题,研究团队提出一种人在环路的故障修复机制。该方案自动挖掘RL策略反复失败的抓取案例,通过VR应用由人类直接演示抓取位置并作为双抓点数据回流训练。仿真实验表明,RL策略已将Contact-GraspNet基线在未见物体上的成功率从55%提升至83%;在此基础上引入58条人类针对性示范后,RL原本持续失败的硬样本成功率从24%跃升至45%,同时保持了整体泛化能力。
技术架构上,人类仅贡献“在哪里抓”的几何先验,机器人仍自主规划接近路径,实现了低带宽高效纠错。该方法解决了纯RL优化平均指标时遗留的长尾失败问题,验证了在点云抓取任务中,少量高质量人类几何反馈能有效修补策略盲区。相比全量遥操作示教,这种仅标注抓取点的VR交互模式大幅降低了数据采集成本,为具身智能算法在复杂非结构化环境中的鲁棒性提升提供了可复现的数据增强范式。
关键看点
- RL性能基线:强化学习将点云抓取策略在未见物体上的成功率从Contact-GraspNet的55%显著提升至83%
- 硬样本修复:仅增加58条人类VR抓取点示范,使RL持续失败案例的成功率从24%翻倍至45%
- 人机协作机制:人类仅在VR中标注抓取位置,机器人自主规划接近轨迹,实现低带宽几何先验注入
- 泛化保持:针对性人类纠错不仅恢复同类未演示案例性能,且未降低对全新物体的抓取成功率
工程与产业观察
该工作揭示了具身抓取算法从“平均可用”迈向“长尾可靠”的关键路径:在RL达到性能瓶颈后,利用VR低成本采集人类几何直觉数据是解决特定失效模式的高效手段。对于仓储与农业等非标物体抓取场景,这种“RL打底+人类定点纠错”的数据飞轮模式,比单纯扩大RL训练规模更具工程性价比,也为构建高价值难例数据集提供了标准化方法。
英文摘要 (Abstract)
Robots in homes, warehouses and farms must pick up objects they have never seen, usually without a 3D model or a category label. A grasping policy that works directly on the point cloud from a depth sensor needs only the geometry the robot observes, so it applies to any new object. Offline grasp generators such as Contact-GraspNet work this way, but they rank grasps by learned plausibility rather than by whether the grasp actually lifts the object. Reinforcement learning (RL) can optimize for that directly and make a point-cloud grasping policy much better on average, yet it leaves specific grasps persistently unsolved. We present a human-in-the-loop failure loop that targets those grasps: situations an RL policy keeps failing are mined automatically, sent to a virtual-reality application where a person picks the object up with their own hands, and returned to training as two grasp points. The person contributes only where to grasp; the robot chooses its own approach. In simulation, RL raises success on unseen objects from 55% with Contact-GraspNet to 83%. Adding 58 human demonstrations nearly doubles success on the hard cases RL kept failing, from 24% to 45%, and recovers performance that RL alone loses on similar cases that were never demonstrated, without reducing success on unseen objects.
Hugging Face Daily Papers 热度:2 upvotes 作者:Achyuthan Unni Krishnan, Karthik Rengarajan, Bharath Narasimhan, Chris von Csefalvay
外部来源
本页事实对应的公开页面,以原文为准。(点击将离开本站)