Nvidia has updated its simulation framework, allowing the Unitree Go1 quadruped robot to achieve walking speeds of 2.0 m/s.
The "sim-to-real gap" is the biggest hurdle in robotics—that is, the problem where a robot trained in a virtual environment trips the moment it hits real, uneven pavement. Using Isaac Sim and Isaac Lab, this new method trains the robot in a digital world to handle obstacles before it ever steps outside. It learns to recover its balance from massive nudges, making it significantly more stable.
Previously, quadrupeds relied on built-in factory controllers that were stable but struggled with sudden chaos. This new approach achieves the same 2.0 m/s velocity and 1.8 rad/s angular velocity as the stock software while gaining much higher recovery capabilities. This progress brings robots closer to environments where they don't need a human to pick them up every time they hit a bump.
Better software means your future delivery bot won't tumble over a curb.
#robotics #Nvidia #tech
Subscribe
