Robust balance
Stay upright through pushes, slips, and uneven ground. Disturbance recovery trained across billions of randomized physics steps.
LimbBot is the control layer that gives humanoid and legged robots robust balance, dynamic walking, and the ability to carry things while they move. You build the body. We ship the motion.
Hardware-agnostic · Runs on-robot at the edge · Built on NVIDIA Isaac & Jetson
Built for the robot makers shipping the next decade of machines
Balance under a shove. Walking on gravel. Walking and lifting at the same time. It is the hardest problem in robotics, and most teams burn years on it. LimbBot solves it once, for any body.
Stay upright through pushes, slips, and uneven ground. Disturbance recovery trained across billions of randomized physics steps.
Walk, trot, climb, and recover on terrain the robot has never seen, with smooth real-time gait control.
Carry, push, and place objects while moving. The control stack coordinates legs and arms as one system.
Instead of welding a control stack to a single robot, LimbBot adapts to many body types through a short per-robot calibration. New hardware ships in days, not quarters.
body classes: bipeds & quadrupeds
simulated steps per policy
real-time edge control loop
to integrate a new robot
GPU-parallel locomotion RL in Isaac Lab & Isaac Sim.
Real-time on-robot inference via CUDA & TensorRT.
Drop-in control node for existing robot software stacks.
OpenUSD scenes and Cosmos terrain for training diversity.
On-robot whole-body control runtime for balance and locomotion.
Learn moreFine-tune and adapt control models to new hardware and tasks.
Learn moreFall avoidance and force limits for robots that work near people.
Learn more"We can build a great robot. Robust whole-body control was the part that kept slipping. LimbBot let us focus on the body and trust the motion."
No. LimbBot is hardware-agnostic. The same foundation model adapts to bipeds and quadrupeds from different makers through a short calibration.
On-robot at the edge for real-time control, typically on Jetson-class compute, with optional cloud fine-tuning in LimbBot Studio.
Most robots integrate in days using the calibration routine, versus the months teams spend building control from scratch.
LimbBot Guard adds a safety layer that limits joint forces and avoids falls, which matters for robots working near people.
Tell us about your hardware and we'll map a pilot. Early partners shape the roadmap.
Request access