RoboScience / 机器科学
About
About RoboScience / 机器科学
RoboScience develops Visics, a general-purpose embodied AI model intended to control manipulation across different robot bodies, objects and tasks. Its VLOA architecture combines an embodied world model with a large manipulation model, learning both how physical scenes may evolve and how a robot should act within them. The object-centred design is intended to transfer behaviours among different embodiments without retraining every task from scratch.
Training data comes from internet-scale video and multimodal physics simulation, supported by RoboMirage, the company's differentiable simulation engine for contacts, rigid-body dynamics and varied end effectors. GPU acceleration and a Python interface let researchers generate interactions and incorporate new objects or grippers into the pipeline. RoboScience was founded in December 2024 and is developing an integrated hardware-software stack around Visics. Current maturity is centred on model development, technical demonstrations and early financing rather than a broadly deployed commercial robot fleet.
Products
- REX G1
REX G1
Funding
- 2026-05-06 - Series A - $146.4M - RoboScience raises RMB 1B Series A
RoboScience announced a RMB 1 billion Series A financing; investors were not disclosed. - source: en.roboscience.co
- 2026-02-12 - Pre-A - RoboScience raises pre-A round of several hundred million RMB
RoboScience announced a pre-A financing of several hundred million renminbi led by Puhua Capital. - source: en.roboscience.co
- 2025-07-30 - Angel - RoboScience raises nearly RMB 200M angel round
RoboScience announced an angel round of nearly RMB 200 million led by JD.com. - source: roboscience.co
Investors
- JD.com - Lead - Unknown - 1 funding event - first seen 2025-07-30
- Puhua Capital - Lead - VC - 1 funding event - first seen 2026-02-12
Team
- Tian Ye - Founder and CEO - He founded RoboScience to build cross-embodiment operating systems and general-purpose intelligence products for robots.
- Shao Lin - Founder and Chief Scientist - As founder and chief scientist, he leads RoboScience's general-purpose manipulation models, physics simulation and cross-embodiment research.
- Penghai Liu - Co-founder - He brings large-scale robot product development, manufacturing and supply-chain execution to RoboScience.
- Tao Wang - Co-founder and Executive President - As co-founder and executive president, he leads capital formation and corporate execution for RoboScience's embodied-intelligence platform and robot deployment.
Timeline
Events
- 2026-08-17 - Product Launch - RoboScience introduces REX G1 mobile manipulation robot
Visics separates a world model that predicts object trajectories from a manipulation model that generates robot control signals. The announced hardware was 540 mm wide, with a 10 kg combined arm holding load and hot-swappable batteries.
Source - 2026-06-24 - Product Launch - RoboScience unveils proprietary Visics embodied-AI model
Visics uses a VLOA architecture combining an embodied world model and a general manipulation model. The launch demonstrated manipulation tasks including furniture assembly, dexterous grasping and production-line work. Its object-centred architecture is designed to generalize across robot embodiments, objects and tasks.
Source - 2026-05-06 - Funding - RoboScience raises RMB 1B Series A
The round amount was RMB 1 billion. The financing followed RoboScience’s February 2026 pre-A round and accompanied development of its Visics general-purpose embodied-AI model.
Source - 2026-02-12 - Funding - RoboScience raises pre-A round of several hundred million RMB
Puhua Capital led the round, joined by Dacheng Capital, Changshi Capital, Hong Kong WiseTech Innovation and Tianqi Capital. RoboScience will use the funding to refine its VLOA technology, which combines embodied world and manipulation models for general-purpose robots.
Source - 2025-09-03 - Product Launch - RoboScience launches RoboMirage simulation platform
RoboMirage supports simulation-based embodied-AI research and training. The platform models contacts across rigid bodies, deformable objects, articulated structures and varied robot end effectors. Its architecture combines high-precision multibody dynamics, extensible contact models, a Python interface and GPU acceleration.
Source - 2025-07-30 - Funding - RoboScience raises nearly RMB 200M angel round
JD.com led the round. China Merchants Venture Capital and SenseTime Guoxiang Capital participated; 01VC returned.
Source
Context
Comparable profiles
- General Trajectory
comparable profile - General AI Robot Learning Lab
- ACE Robotics (大晓机器人)
stack adjacency - Embodied AI Models · Robotics Foundation Model Labs
- ForgeAI
stack adjacency - Embodied AI Models
- Alibaba Group
stack adjacency - Robotics Foundation Model Labs
- One Robot
stack adjacency - Robotics Foundation Model Labs · Robot Learning
- Mbodi AI
stack adjacency - Robotics Foundation Model Labs
- The Robot Learning Company
stack adjacency - Robot Learning
- ATOM Humanoid
stack adjacency - Robotics Foundation Model Labs
Shared workflows
- Physical Intelligence
workflow adjacency - Cross Embodiment Robot Learning
- Intelligence Factory
workflow adjacency - Cross Embodiment Robot Learning
- Algomatic Dynamics
workflow adjacency - Cross Embodiment Robot Learning
- Reward AI
workflow adjacency - Cross Embodiment Robot Learning
- Lingsheng Technology
workflow adjacency - Cross Embodiment Robot Learning
- Dream Labs
workflow adjacency - Cross Embodiment Robot Learning
Context tags
- Embodied AI - Capability - Company develops embodied-AI model systems.
- Robot learning - Capability - Cross-embodiment manipulation training
- Cross-embodiment robot learning - Workflow - Visics is positioned for cross-embodiment robot manipulation.
- General AI robot learning lab - Peer Group - Company develops general robot-learning models.
- Embodied AI models - Stack Layer - Visics embodied-AI model
- Robotics foundation-model ecosystem - Stack Layer - Visics is an embodied-AI model for robot manipulation.