Artificial intelligence news.
Straight from the original announcements.

Motphys launches MotrixSim 1.0 for embodied AI robot training

November 14, 2026, 10:49By the AI News Bites Editorial TeamSource: PR Newswire

Chinese company Motphys has launched MotrixSim 1.0, a fully in-house physics engine designed to support embodied AI simulation and robot training. The company showcased the system at CoRL 2026 and highlighted results from tests involving grasping, simulation speed, perception and complex mechanisms.

Key points
  • Motphys launched its fully in-house MotrixSim 1.0 physics engine.
  • The engine achieved a 100% pass rate in the company's grasping test at 2 ms and 10 ms.
  • Motphys reported 10.2 minutes for a G1 tracking task, versus 130.8 minutes for an Isaac Sim-based solution.
  • MotrixSim supports simulation across 4,096 parallel environments.
  • Forge adds a Blender-based tool for AI Agents to generate physics scenes.

An in-house engine for embodied AI

Motphys says it developed MotrixSim 1.0 from the ground up in Rust rather than building on existing open-source physics engines. According to the company, this approach provides strong memory safety and computation power while allowing the system to compete with established international platforms at the underlying architecture level.

The company described a shift in the role of simulation as robots become more dependent on large amounts of interaction experience. "Simulation once served to validate how machines perform. Now, it is where they learn how to perform," Motphys co-founder and CTO Zhang Heng said, according to the PR Newswire announcement.

Motphys said simulation is evolving from a post-design verification tool into a system for producing physical experience that can be used in embodied AI training.

Tests cover grasping, speed and complex motion

In a disturbance-rejection grasping test using a Franka arm and objects with different shapes, Motphys said MotrixSim was the only engine to achieve a 100% pass rate at both 2 ms and 10 ms simulation step sizes. In a critical friction test, the company said the engine reproduced hold-and-slip behavior near the critical friction point in line with Coulomb's law, with 0mm slip distance.

MotrixSim supports CPU and GPU parallel computation. Motphys said it can deliver millions of physics steps per second across 4,096 parallel environments. In a controlled G1 whole-body tracking comparison using the same task configuration and reinforcement learning algorithm, MotrixSim 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 company described the difference as about 12.8×.

For a Strandbeest mechanism with 92 linkages, Motphys said its solver maintained the assembly relationships at a 5ms time step by solving closed-loop constraints together with multibody dynamics in generalized coordinates.

Tools for perception and deployment

Motphys also reported a LiDAR simulation throughput of 124 million rays per second. It said visual synthesis maintained stable throughput when resolution increased 16 times.

For developers, MotrixSim offers high-speed solving through a single line of code, which Motphys said achieved a 98.42% success rate in its testing. The company also highlighted a "zero-conversion + AI auto-modeling" workflow.

Motphys launched Forge alongside MotrixSim 1.0. The company described Forge as the industry's first Blender-based extension tool for enabling AI Agents to automatically generate physics scenes. Motphys said its research has been accepted to conferences including RSS and CoRL, and that it has built a simulation and real-robot training center for embodied AI in China.

The launch positions MotrixSim alongside the physics engines Motphys identifies as leading systems in the field, including Google's MuJoCo and NVIDIA's Isaac Sim. For robot developers, the company's stated goal is to make physical experience available as digital training data that can be generated at scale, potentially reducing reliance on slow and expensive real-world trial and error.

MotphysMotrixSim 1.0CoRL 2026FrankaIsaac SimG1roboticsphysical AIbenchmarks

Related news

Anyscale uses Ray to cut JEPA robot planning below one second
For developersNovember 12, 2026, 10:00

Anyscale uses Ray to cut JEPA robot planning below one second

Anyscale has published a design and tested implementation that moves V-JEPA2-AC world model inference into a Ray cluster, reducing planning time to 0.86 seconds per…

ModelsTop story
ModelsNovember 9, 2026, 20:51

Google DeepMind unveils Gemini Robotics 2.5 for more capable robots

Google DeepMind has introduced Gemini Robotics 2.5, a vision-language-action model designed to help robots learn a broader range of skills and operate more reliably in…

FFR Says OmniMind Tops RoboColiseum Embodied AI Ranking
ResearchNovember 9, 2026, 02:42

FFR Says OmniMind Tops RoboColiseum Embodied AI Ranking

FF EAI Robotics Ecosystem says its OmniMind embodied AI model ranked No. 1 on RoboColiseum's global leaderboard as of November 6. The company also reported sales…