Motphys Claims Robot Training 12.8x Faster

Summary

Pixel shares Motphys's claims about MotrixSim 1.0, a Rust-built robot simulation engine unveiled at CoRL 2026. In company-run tests, it reached a robot tracking score in 10.2 minutes versus 130.8 minutes for an Nvidia Isaac Sim-based setup, raising questions about faster robotics training and benchmark comparisons.

DEV Community

China's Motphys Says It Trains Robots 12 Times Faster Than Nvidia

Chinese company Motphys said its new MotrixSim 1.0 robot simulation engine reached a training target about 12.8 times faster than a setup built on Nvidia's Isaac Sim, based on its own tests. Key Takeaways Motphys showed MotrixSim 1.0 at CoRL 2026 as a fully in-house embodied intelligence simulation engine Motphys says MotrixSim reached the current tracking score in about 10.2 minutes, versus 130.8 minutes for an Isaac Sim-based solution Motphys says MotrixSim was the only engine with a 100% pass rate at both 2 ms and 10 ms step sizes in a Franka arm grasping test Motphys says it built the engine core in Rust rather than on existing open-source engines Motphys announced on Nov. 14 that it was invited to attend CoRL 2026, where it showed MotrixSim 1.0. All performance figures in this story come from the company's own release. Also Read: Chef Robotics Moves Food-Prep Arms From Pilot Kitchens to Live Lines on Nvidia Isaac Built From Scratch In Rust Motphys says the physics engine race has long been dominated by Western players. The release names Google's MuJoCo and Nvidia's Isaac Sim as the world's leading physics engines. The company says it developed the MotrixSim core from the ground up in Rust. It says this gives the engine strong memory safety and computation power. Co-founder and CTO Zhang Heng said, "Simulation once served to validate how machines perform. Now, it is where they learn how to perform." The Benchmark Claims From Motphys In a G1 whole-body tracking task, Motphys says it used the same task environment configuration and the same RL algorithm for both setups. It says MotrixSim reached the current tracking score in about 10.2 minutes of end-to-end time. The Isaac Sim-based solution took 130.8 minutes, a gap of about 12.8 times. In a disturbance-rejection grasping test with a Franka arm and objects of different shapes, the company says MotrixSim 1.0 was the only engine to reach a 100% pass rate at both 2 ms and 10 ms step sizes. In a critical friction test, it says the engine showed hold-and-slip behavior consistent with Coulomb's law and 0mm slip distance. Motphys also says traditional engines fail to keep the Strandbeest mechanism, which has 92 linkages, stable at a 5ms step. It says MotrixSim keeps the mechanism's assembly relationships during motion. Metric Figure reported by Motphys G1 tracking time, MotrixSim About 10.2 minutes G1 tracking time, Isaac Sim-based 130.8 minutes Franka grasp pass rate 100% at 2 ms and 10 ms Parallel environments 4,096 LiDAR throughput 124 million rays per second Single-line solving success rate 98.42% Compute, Perception And Developer Tools Motphys says a pipeline supporting CPU and GPU parallel computation lets MotrixSim reach millions of physics steps per second across 4,096 parallel environments. It says visual synthesis keeps stable throughput even when resolution rises 16 times. For developers, the company cites a "zero-conversion + AI auto-modeling" workflow. It also launched Forge, which it calls the industry's first Blender-based extension tool, enabling AI Agents to automatically generate physics scenes. Motphys says its research has been accepted to conferences such as RSS and CoRL. It says it has built a simulation and real-robot training center for embodied AI in China. Read Next: RoboScript Teaches Robots A Task From One Demo, Beating Trained Rivals By 20 Points Originally published on Fathom News.
Motphys Claims Robot Training 12.8x Faster
Powered by Readybot.io

The latest from Pixel

Incident Response: Look Beyond the Red Alert

Pixel’s DEV Community announcement explores how teams can fixate on one failing service while overlooking the real cause of an outage, such as a full …

Sach: AI Fact-Checks WhatsApp Forwards in India

Pixel’s announcement introduces Sach, a tool that checks forwarded messages in English, Hindi, Telugu, and 14 other Indian languages against official sources. It gives cited …

Java 25 Removes SecurityManager: What to Fix

Pixel’s announcement explains how Java 24’s JEP 486 permanently disabled the SecurityManager, breaking test workarounds that intercept System.exit. It recommends returning exit codes or throwing …