Website · Research · Hugging Face · X · Contact
MirroS is building Physical Recursive Self-Improvement (Physical RSI): intelligent systems that transform every encounter with the unknown into verified, reusable capabilities.
We mirror reality into an evolving, agentic world model—a computational representation that can be executed, tested, revised, and improved. Through interaction and feedback, the world model and the agent co-evolve: each surprise reveals what the system does not yet understand, expands the environments it can reason about, and drives the next improvement in how it acts.
Research Read Our Thesis ↗
Our roadmap spans environment understanding and abstraction, agentic physical reasoning, continual learning from experience, and discovering the unknown. Together, these capabilities form the recursive substrate for intelligence that grows beyond a fixed problem space.
Our team brings together talent from Tsinghua University and Peking University, with industry experience across NVIDIA, Tencent Hunyuan, Kling AI, and ByteDance Seed. Our expertise spans world models, multimodal learning, physical reasoning, agentic systems, reinforcement learning, robotics, and embodied intelligence.
We are actively recruiting researchers and engineers across disciplines. If you are passionate about building self-improving intelligence for the open physical world, reach out at [email protected].