Packbot Active Weight Shifting
We applied Dynamic Threshold Learning to learn how to actively shift the center-of-gravity of the PackBot using its manipulator arm. In these experiments, DTL learns the arm angle (action) needed to prevent rollover as a function of pitch angle (state). Then, as an operator drives the robot forward, DTL automatically moves the arm to shift the robot's center-of-gravity and prevents rollover as the robot climbs a tall obstacle. Using this behavior, the robot is able to climb an obstacle that it would not otherwise have been able to climb. This video shows DTL learning how to move the robot's arm to allow it to climb a tall obstacle:
After a few learning trials, DTL is able to successfully manipulate the robot's center-of-gravity to allow the robot to climb over the obstacle. Additional details are available in our paper submitted to ICRA 2012.
