BEV 杀入具身智能：跨维把机器人数据带上 Scaling 快车道
英文摘要
Chinese embodied intelligence company/research group Kuawei has adapted Bird's Eye View (BEV) techniques, originally from autonomous driving, for robot data representation. The method aims to unify spatial perception, enabling large-scale training of robotic systems. By applying BEV, robot learning data can be scaled efficiently, similar to the scaling laws observed in large language models. This reflects a trend of cross-domain technology transfer from autonomous vehicles to general-purpose robotics. Limited technical details are provided in the announcement.
中文摘要
中国具身智能公司/研究团队跨维将自动驾驶领域的鸟瞰图（BEV）技术引入机器人数据表示。该方法旨在统一空间感知，实现机器人系统的大规模训练。通过应用 BEV，机器人学习数据可以高效扩展，类似于大语言模型中的扩展定律。这体现了自动驾驶技术向通用机器人领域的跨域迁移趋势。该公告未透露详细技术细节。
关键要点
Kuawei introduced BEV-based spatial representation to embodied intelligence, unifying robot perception.
跨维将鸟瞰图（BEV）空间表示引入具身智能，统一机器人感知。
The approach enables scalable robot data training, potentially unlocking scaling laws for robot learning.
该方法使机器人数据训练可扩展，有望解锁机器人学习的扩展定律。