Let’s talk

Building the missing physics
layer of robotic simulation.

Actuator heat loss, battery life, magnetic fields and more, in your robotics simulation stack.

TODAYJoint angle, velocity, torque
WITH INVISIBLE FORCEActuator heat, fields, battery
Scroll

Unitree G1 · Illustrative animation

Rigid bodies are not enough

Today’s robotics simulators model rigid bodies. A motor is reduced to stiffness and damping, and the physics inside the hardware is left out.

We add the missing physics:

  • Actuator heating: Joule losses against rated current, and the thermal stops they cause
  • Battery management: temperature, state of charge, current and voltage limits
  • Magnetic grippers: field and holding force. Isaac Sim has no magnetic model.
  • Microgravity: for robotic maintenance of orbital datacentres
  • Tactile and piezoelectric sensing: for dexterous hands, where most teams still have no simulation

We couple existing rigid-body simulators to multiphysics solvers, on the GPU. You keep the scale you need for reinforcement learning and world models, and gain the parameters that close the sim2real gap.

Demos

RECORDED IN ISAAC SIM
01

Magnetic gripper

Magnetic field and holding force of a permanent-magnet gripper, computed on the GPU. Next: the SCHUNK EMH RP 084.

ILLUSTRATIVE ANIMATION
02

Actuator Joule heating

A robot’s joints change colour as they heat up under load and cool down at rest.

IN DEVELOPMENT · ILLUSTRATIVE VALUES
03

Simulated battery management system

A Unitree G1 walks while its joint torques drive battery current and state of charge in real time.

Render of the Canadarm on the Space Shuttle in orbitCANADARM · RENDER
04

Microgravity for space robotics

The Canadarm, with real mass properties, capturing a free-floating payload in zero gravity.

Optimized computing

GPU-accelerated simulation

Running on NVIDIA Warp, a purpose-built open-source Python framework that delivers GPU acceleration for computational physics, AI and optimization workflows.

AI-accelerated physics

Combining CFD and FEA solvers with machine learning to make traditional physics simulation much faster.

Sim2real for autonomous, 24/7 deployment

Most simulations run short episodes in isolated environments. Real robots run autonomously for hours, and small effects add up: heat builds in the joints, cells drift apart, charge runs out.

Simulating the full mission shows the forced stop before it happens in the field, and lets you validate the control strategy that avoids it: fewer motor starts and brakes, less energy.

VLAs go from prompt to pixels to actuator commands, with little control in between. Physical limits like actuator temperature give them a safety guardrail.

A Unitree G1 humanoid in a blue jacket on a snowy summit above the clouds
ROBOT EVEREST ↗APRIL 2027

Testing the model on Everest

We will test the simulation in the harshest environment on Earth. As part of Robot Everest, we will validate sim2real on a real Unitree G1 attempting to become the first humanoid to climb in the Himalayas, in April 2027.

Cold, altitude and a full autonomous mission will give us motor, battery and thermal data no lab can produce.

Photo: Robot Everest