Deploying an RL model in the Spot robot from Boston Dynamics
In this section, we will see a sample project that trains, tests, and deploys an RL model in Boston Dynamics’ Spot robot. Figure 15.12 shows the training pipeline using Isaac Lab and Isaac Sim.
Figure 15.12: Training of Spot robot in simulation for locomotion (source: https://developer.nvidia.com/)
The task they are trying to do is robot locomotion. If we give the velocity command, the robot should move based on it.
After training on the locomotion task in Isaac Sim, Figure 15.13 shows the testing and deployment of the model on actual hardware.
Figure 15.13: Testing of the trained model on actual Spot robot (source: https://developer.nvidia.com/)
They used Jetson Orin as the computer and read the sensor state using their APIs. We can also use ROS 2 or Isaac ROS to get the robot’s sensor state. The command to the robot is the velocity command from the user using a joystick. The actions...
