Massachusetts Institute of Technology Cambridge United States
This report describes the methods, procedures and results of developing software infrastructure for allowing remote control of a bipedal human-scale Atlas robot. This work focuses on allowing remote operations in severely communications restricted environments for search and rescue. The research relied heavily on abstraction methods for issuing commands, compression and processing of sensor data, and ensuring the safety of the robot during execution. The DARPA Robotics Challenge provided a live demonstration to showcase the robots performance in a simulated search and rescue environment. The abstraction techniques developed from both a control and terrain estimation viewpoint, were successfully able to allow autonomous operation. These techniques substantially reduced the amount of communication necessary between the human operator and the robot allowing them to operate with substantial delays in communication as well as complete losses of communication. The software developed was successfully able to control the robot to complete a significant majority of tasks to drive a vehicle, dismount the vehicle, open a door, manipulate various objects and ultimately walk through uneven surfaces and climb stairs. MIT placed sixth in the finals which was an impressive finish given that their best run required them to operate without the use of one arm which was damaged in a fall.