Adaptive user interface for vehicle swarm control
Paul Robinette · 2010
An algorithm to automatically generate behaviors for robotic vehicles has been created and tested in a laboratory setting. This system is designed to be applied in situations where a large number of robotic vehicles must be controlled by a single operator. The system learns what behaviors the operator typically issues and offers these behaviors to the operator in future missions. This algorithm uses the symbolic clustering method Gram-ART to generate these behaviors. Gram-ART has been shown to be successful at clustering such standard symbolic problems as the mushroom dataset and the Unix commands dataset. The algorithm was tested by having users complete exploration and tracking missions. Users were brought in for two sessions of testing. In the first session, they familiarized themselves with the testing interface and generated training information for Gram-ART. In the second session, the users ran missions with and without the generated behaviors to determine what effect the generated behaviors had on the users' performance. Through these human tests, missions with generated behaviors enabled are shown to have reduced operator workload over those without. Missions with generated behaviors required fewer button presses than those without while maintaining a similar or greater level of mission success. Users also responded positively in a survey after the second session. Most users' responses indicated that the generated behaviors increased their ability to complete the missions.