How it works

A vectorized simulator runs many training environments in parallel. A structured reward guides the robot toward useful navigation behavior and enables policies to generalize.