RoboScience raises several hundred million yuan for robot intelligence
Develops AI models, a physics simulator and wheeled robots that learn how to handle objects. Its goal is to reuse learned manipulation skills across different robots and tasks, rather than programming each movement separately. RoboScience announced a Series A2 of several hundred million yuan on November 2, without an exact amount or valuation. Funding supports model development, application validation and hiring. Its proprietary Visics model plans object motion before converting it into commands for different robot hardware; RoboMirage generates simulated training data. Current logistics, retail and industrial work includes proof-of-concept validation, not confirmed mass deployment.
DTT summary based on 21jingji / PEdaily's report. The full article is on the publisher's site.
Additional reporting
A Series A2 round of several hundred million yuan announced November 2, 2026. The exact amount and valuation are not disclosed in the checked reports.
琥珀资本 (Hupo Capital), 鸿瑞达投资 (Hongruida Investment) and 上海半导体产投 (Shanghai Semiconductor Industry Investment). The original Chinese names are retained to avoid uncertain English brand translations.
Confirmed across the checked reports: 爽资本 (Shuang Capital), 360 strategic investment and 金鼎资本 (Jinding Capital), plus several listed-company investment platforms. Reports disagree on whether another participant is Guolian Tongbao or Guolian Minsheng Capital; that investor is not treated as verified here.
Visics, a general-purpose embodied AI model; RoboMirage, a physics simulation engine; and REX G1, a wheeled humanoid robot. Embodied AI means software that controls actions in the physical world, not only text or images.
Visics uses Vision-Language-Object-Action, or VLOA. It first predicts how objects should move, then translates those trajectories into control signals for the particular robot. This separates high-level task planning from hardware-specific movement.
RoboMirage simulates contact and object motion to generate training examples. The company combines simulated trajectories with automatically labelled video. Data-scale and cost-reduction figures in company materials are company claims, not independent benchmark results.
The company is validating applications in logistics, retail and industrial settings. Visics is offered through Tencent Cloud interfaces, alongside robot controllers and full-system products. Proof-of-concept work is not presented as completed large-scale customer deployment.
Iterate the embodied AI model, run proof-of-concept validation and commercial deployment work in logistics, retail and industrial settings, and recruit technical staff.
Tian Ye, CEO, and Shao Lin, chief scientist, founded RoboScience in December 2024. English ordering of names follows the company team page.
The company disclosed a pre-A round of several hundred million yuan in February 2026 and lists an RMB1 billion A-round milestone in June 2026. Those earlier financings are separate from the November A2 announcement; no cumulative total is inferred.
