MotrixLab · Imitation & Reinforcement Learning Framework
Train robot policies in simulation and deploy them to real robots, across body types from humanoids and quadrupeds to arms and dexterous hands, supporting tasks from locomotion to fine manipulation.
One framework, from basic locomotion to whole-body control
One training pipeline works across humanoids, quadrupeds, arms and dexterous hands — handling whole-body motions like rugged-terrain walking, backflips and wall flips, and manipulation tasks like grasping, cabinet opening and in-hand control.
Arm · Lift Cube
Franka
Policies trained in simulation, deployed straight to real robots
Backed by MotrixSim's physical accuracy, trained policies have been validated on real humanoids, quadrupeds and dexterous hands.
Sampling overlaps learning — faster training on the same hardware
Simulation sampling and policy learning overlap within the same cycle without waiting on each other, cutting end-to-end training time significantly.
Quadruped training case
3–10×
End-to-end speedup on same hardware
12s
Train one quadruped gait (measured)
202K
steps/s · 12,288 parallel envs
Test setup · RTX 4090 + Ryzen 9 9950X3D
Layered architecture, environment decoupled from training
Training environments and training logic are split into layers, making the framework easy to extend.
Layered architecture
Interface Layer
Algorithm Layer
Environment Layer
Physics Layer
MotrixSim · In-house EngineDesign advantages
- Decoupled modules
- — Environment development fully separated from training logic
- Flexible config
- — Layered configuration with runtime overrides
- Extensible
- — Add new components easily via the registry system
- Multi-backend
- — One environment runs on different simulation and training backends
- Experiment-friendly
- — Configs can be saved and compared for reproducibility
Community vs Commercial
MotrixLab offers a free community edition for academic research and learning; the commercial edition is for commercial use and delivery.
COMMUNITY
CommunityFree
- Free & open — academic research & learning only (no commercial use)
- Full framework
- Locomotion training framework & sample projects
- Community & docs support
- Tracks public releases
- Commercial use, redistribution & embedding
- Adds manipulation & imitation-learning frameworks
- On-premise deployment & cloud service
- Domestic CPU / GPU / NPU support
- Official engineering support + SLA
- Co-development & scene buildingContact us
- Training & consulting
- Contract maintenance, upgrades & renewal benefits
Items marked “Add-on” are purchased and priced separately on top of the commercial license; “Contact us” items are negotiated case by case. Students and faculty may apply for an academic license of the commercial edition based on their research.