RBR50 recap: Spotlight on robotics innovation

Engineering & simulationNvidia Isaac Lab20% more energy efficient

Agility Robotics used Nvidia's Isaac Lab simulation environment to train a reinforcement learning model from scratch, allowing the Digit robot to learn how to walk and pick objects without prior human demonstration, resulting in a 20% improvement in energy efficiency compared to previous methods.

“They really had started this project like 2 weeks before GTC to train, let, uh, digit train itself to walk from scratch in simulation. … the system learned to walk on its own … it was 20% more efficient from an energy perspective than what they had developed for the last decade before this.”

Engineering & simulationNvidia Isaac Lab

Boston Dynamics used Nvidia's simulation tools to transfer motion capture data to their Atlas humanoid, enabling the robot to perform complex movements like cartwheels and dancing using reinforcement learning.

“that was all done with a human or a motion capture suit, and then they were able to transfer that motion onto the robot using reinforcement learning, which is wild.”

Engineering & simulationNvidia Isaac Lab

Nvidia enabled developers to use Isaac Lab to randomly generate diverse elements and environments in simulation, accelerating the robot learning process through massive parallelization to reduce the sim-to-real gap.

“one of the cool things that Invinia does is it enables some relation to add, uh, to randomly generate these different elements or environments, which, again, accelerates the whole learning process that, uh, instead of having, imagine a floor with 5 robots trying to do something again and again and something in it right. You can do that on a massively parallel scale.”