Skip to content
AI.info

The Pulse

Motphys launches MotrixSim 1.0 for embodied AI simulation

Motphys showcased its in-house physics simulation engine at CoRL 2026. The company says a training comparison favored its software over an Isaac Sim-based solution.

Motphys launches MotrixSim 1.0 for embodied AI simulation

Motphys has launched MotrixSim 1.0, an in-house physics simulation engine for embodied AI and robot training. The Chinese company says it showcased the software at CoRL 2026 and reports that it outperformed an Isaac Sim-based solution in a training comparison.

Source prnewswire.com

The engine was developed from the ground up in Rust rather than built on existing open-source engines, according to Motphys. Its product page describes MotrixSim as a platform for training robot policies and generating synthetic data.

Training comparison

In a controlled comparison on a G1 whole-body tracking task, Motphys says its simulator reached the current tracking score in about 10.2 minutes of end-to-end time, compared with 130.8 minutes for an Isaac Sim-based solution. The comparison used the same task environment configuration and reinforcement-learning algorithm.

The company puts the gap at about 12.8 times. It also says the engine can reach millions of physics simulation steps per second across 4,096 parallel environments using a pipeline that supports parallel computation on CPUs and GPUs.

Motphys co-founder and chief technology officer Zhang Heng described a change in simulation’s role: “Simulation once served to validate how machines perform. Now, it is where they learn how to perform.” The company’s stated aim is to turn real-world physical experience into digital learning resources that can be generated at scale.

Grasping and motion tests

In a disturbance-rejection test, a Franka arm grasped objects of different shapes. Motphys says its engine was the only one to achieve a 100% pass rate at both 2 ms and 10 ms simulation step sizes.

In a separate friction test, the company reports that the simulator showed hold-and-slip behavior near the critical friction point consistent with Coulomb’s law, with a slip distance of 0 mm.

Motphys also describes a test of a Strandbeest closed-loop mechanism with 92 linkages. At a 5 ms time step, Motphys says its engine maintained the mechanism’s assembly relationships during motion by solving closed-loop constraints together with multibody dynamics in generalized coordinates; the company says traditional engines failed to maintain the mechanism’s stability at that step size.

Scenes and licensing

Motphys reports LiDAR simulation throughput of 124 million rays per second. It says visual synthesis throughput stayed stable when resolution increased 16 times.

The company also launched Forge, a Blender-based extension intended to let AI agents automatically generate physics scenes. Motphys describes a workflow involving no conversion and automatic AI modeling.

Motphys offers a free community edition of the platform for academic research and learning, but not commercial use. Its product page lists built-in robot bodies, basic terrain and objects, model imports and community documentation among the community features.

GPU acceleration is listed under the commercial edition, which also permits commercial use, redistribution and embedding. Custom bodies and scene assets, as well as certain import and scene-reconstruction options, are separately priced add-ons. Students and faculty can apply for an academic license of the commercial edition based on their research.

Sources

Related stories

Explore

More articles