Motion Brain
Shanghai-based embodied intelligence company that develops an on-device, generative, cross-embodiment robot 'brain' built on its MotionGPT series of motion foundation models.
- CEO / Founder
- Mu Zelin (穆泽林)
- Stage
- Early Stage
- Total Funding
- $126M
- Latest Round
- Series A
- Key Investors
- Shanghai Guohe Investment; Furong Investment (Fudan); Xuhui Capital; Green Technology Capital (绿技行资本); Fujian Provincial Investment Capital (闽招基金); Leo Group; Huafu Capital; Yingjue Glory; Gengxin Capital; Chuanghehui Capital; Jinyue Investment; Shenbao Yiben Fund; Orient Securities; Shaanxi High-Tech Industry Investment; Anyu Fund; Tianmeng Investment; Jianyuan Tianhua
Technology & Products
Key Products
Motion Brain general embodied brain; MotionGPT, MotionChain, MotionStream and MotionGPT3 generative motion models; LL3DA point-cloud 3D understanding, reasoning and decision model; T²MB embodied large model (March 2026); STI-WM spatio-temporal world motion model (2026); compressed embodied models for on-device inference; virtual-real fusion training ground solutions.
Technological Advantage
Team spun out of Fudan University's School of Future Information Innovation. Research record includes an IJCAI 2025 Best Paper for embodied model compression (80%+ parameter reduction), an ICCV 3D object recognition challenge win and a first place in CVPR 2024 3D dense captioning. MotionGPT3 posted top results on the HumanML3D benchmark. As of March 2026 the company reported adaptation to 10+ robot bodies (Unitree G1, Qinglong, Dobot, Realman and others) and 5+ edge chip families (NVIDIA Orin/Thor, Horizon J5/J6, Huawei Ascend 310B).
Differentiation
Value Proposition
A general-purpose robot brain under a 'one brain, many forms' approach. The company states it can deploy the brain to a new robot body in 2-3 weeks and that it runs on edge chips from NVIDIA, Huawei Ascend and Horizon Robotics. The stated aim is to cut the cost of re-developing robot intelligence for each body and industry.
How They Differentiate
Builds a hardware-agnostic, generative, on-device brain rather than its own robot body. It models motion as about 3,000 motion primitives predicted as tokens (a 'world motion model' approach). It emphasizes model compression and native support for domestic Chinese compute chips. Investors describe its positioning as an 'Android for robots'.
Market & Competition
Target Customers
Robot hardware makers (humanoid, cleaning, logistics, industrial), property management and sanitation service companies, retail groups, home-appliance factories, elderly-care operators, and embodied-AI training and data-collection centers.
Industry Verticals
Urban life services (property management, sanitation); smart manufacturing (loading/unloading, quality inspection, handling, sorting); smart retail; elderly care; embodied AI training and data-collection centers.
Competitors
Galbot (Galaxy General Robotics); Galaxea (Xinghaitu); AgiBot (Zhiyuan Robotics)
Growth & Milestones
Growth Metrics
Audited cash-collected revenue of about 10M yuan in 2025 and 30M yuan in H1 2026; full-year 2026 revenue projected above 50M yuan. Valuation reportedly increased more than 10x in H1 2026. The Pre-A round was reported as 5x oversubscribed.
Major Milestones
2023: team released MotionGPT. Jan 2025: company founded in Shanghai. Sep 2025: MotionGPT3 released; IJCAI 2025 Best Paper. Dec 2025: angel round led by Shanghai Guohe Investment; top award at the national disruptive technology innovation competition. Mar 2026: T²MB embodied model released. May 2026: ¥300M Pre-A round. Jul 2026: ~¥100M Pre-A extension. Aug 2026: ~¥500M Pre-A+ round.
Notable Customers
Unitree Robotics; HCFA; Xiaomi Group; a listed property management company; an A-share listed sanitation services company