DEEPRobotics-Brain

DEEPRobotics-Brain is the embodied-AI "brain" unveiled by DEEP Robotics (云深处科技) on October 24, 2026, together with five research results spanning world models, embodied foundation models, navigation agents, policy optimization and a data loop.

The five results

• DR-WM-1.0 (world model): a physics-consistent world model for quadruped robots with friction, roughness and stiffness priors, pretrained on 2.58M images + 100k+ real-robot clips (arXiv 2610.17630)
• DR-VLA-2.0: embodied foundation model
• DR-Agent: navigation agent
• DR-APO: policy-optimization framework that lifted NavDP simulation success from 70.33% to 89.55%
• DR-Benchmark: data-loop platform

Collaborators

Co-developed with MIT, UCSD, UIUC, Tsinghua, ZJU, HKUST and BIT.

Coverage

Reported by 新京报 (Beijing News), 环球网 (Huanqiu), ai123 and 机器人大讲堂, among others.

Who should follow it

• Researchers in quadruped/legged robotics and embodied AI
• Developers working on robot world models, VLA models and navigation policies
• Industry watchers tracking Chinese robotics companies' technical roadmaps

Further reading

• Beijing News report: bjnews.com.cn/detail/1792829104129443.html
• World-model paper: arXiv 2610.17630

Figures come from the official announcement and media coverage; refer to the paper and official materials for details.