AMPC Baseline
The final agent's average speed was about 3.33 times the AMPC average speed. The final Infoprop Dyna racing policy substantially outperformed the AMPC baseline. The learned racing agent stayed closer to the track centerline and completed m…
1 sources - 5 claims
The final agent's average speed was about 3.33 times the AMPC average speed. The final Infoprop Dyna racing policy substantially outperformed the AMPC baseline. The learned racing agent stayed closer to the track centerline and completed more than three times as many laps as the AMPC agent. The controlled slipping strategy differed from the AMPC strategy of using gyroscopic effects to reorient the robot. The real-world dataset was initialized with trajectories collected by manually driving the robot using AMPC and a joystick interface.