DISCOVER Robotics Raises $100M Angel-Plus Round: Embodied AI Starts Getting Valued as Full-Stack "Model–Data–Simulation"
Category: industry · embodied AI / fundraising Date: 2026-08-03 08:35 (Beijing Time) Sources: Leifeng exclusive report; DISCOVER Robotics official GitHub and project materials
The Numbers, First
According to a Leifeng report, DISCOVER Robotics completed a $100 million angel-plus funding round on August 3. Participants include IDG, Xinglian, Wu Yuefeng, Dachen, Joyoung, Huaying, Binhu District Industrial Group, among others, with existing shareholders also increasing their stakes. The report also notes that the company's previous angel round of over $100 million closed less than a month earlier.
One important caveat: while the article used the phrase "embodied AI unicorn," no post-money valuation, equity stake, or investment terms have been publicly disclosed. A $100 million raise is not the same as a $100 million valuation, and certainly cannot be treated as an independently audited valuation.
It's Not Selling a Single Robot
DISCOVER splits its product roadmap into three layers:
- ORION: a hierarchical embodied AI foundation model responsible for higher-level perception, decision-making, and action generation;
- LOOP: a hybrid real-virtual data flywheel aimed at connecting simulation, real-robot feedback, and training data;
- DISCOVERSE: an open-source, high-fidelity robotics simulation framework built on MuJoCo and 3D Gaussian Splatting, covering workflows for data collection, imitation learning, diffusion policy, and RDT.
- https://www.163.com/dy/article/L3D54AQV05118HA4.html
- https://github.com/DISCOVER-Robotics/DISCOVERSE
- https://github.com/DISCOVER-Robotics
- https://air-discoverse.github.io/
The DISCOVERSE GitHub repository confirms it is MIT-licensed and a modular Real2Sim2Real-oriented simulation system supporting ACT, Diffusion Policy, RDT, and custom algorithms. The repo also explicitly states that the public README does not provide unified simulation-speed or real-task success-rate benchmarks. In other words, we can confirm the "foundation exists and the interfaces are open," but not that ORION's generalization has reached commercial deployment level.
Why It Matters
The real signal of this round is not "yet another robot company raising big money," but that capital is starting to pay for a complete closed loop: models solve generalization, simulation solves data and long-tail cases, and real-robot feedback closes the reality gap. In embodied AI, the most expensive part is often not training a model once, but continuously obtaining interaction data that is abundant, clean, and reproducible.
If this closed loop truly works, the company's moat will shift from hardware design toward the compounding effect of "data generation → policy training → real-robot recycling." If it doesn't, the three-layer product stack risks becoming three independent projects, and the fundraising figure will remain just a fundraising figure.
Original sources and evidence: