Motphys

Research

Motphys takes an active part in academic research, advancing robot learning together with universities and industry — and welcomes new collaborations.

Publications & open source

All research

Tsinghua × Motphys2026.05

UniLab

A Heterogeneous Architecture for Robot RL Beyond GPU-Dominant Paradigms

UniLab

A heterogeneous robot-RL architecture: CPU-side parallel physics, GPU-side learning, decoupled by a unified runtime. Motphys's MotrixSim serves as the parallel physics backend powering this CPU-sim / GPU-learn path.

3–10×

End-to-end speedup

12s

Quadruped training

RSS 20262026.04

GS-Playground

A High-Throughput Photorealistic Simulator for Vision-Informed Robot Learning

GS-Playground

A high-throughput photorealistic simulator for vision-informed robot learning, built on 3D Gaussian Splatting with minute-scale Real2Sim digital twins and zero-shot visual grasp transfer. Accepted at RSS 2026.

10,000 FPS

Render speed

90%

Zero-shot grasp