This project addresses the control and communications among underwater robotic vehicles to explore and map in ocean environments, where the communications are inherently low bandwidth, may be degraded and even disrupted due to natural ocean phenomena. The research focuses on coordinating robots and cooperating teams of such robots under such conditions with limited human intervention. It is expected the principles learned from this project will be generalizable to the deployment of teams of cooperating robots operating in harsh environments with similar communication challenges such as what might be expected in the aftermath of natural disasters.
The project addresses technical challenges with robots learning to describe their environment using high-level descriptors, using state of the art unsupervised machine learning techniques, and exchanging compact messages with each other to keep the description model consistent across all the cooperating robots. This high-level scene description is used by the robots to efficiently communicate the state of the exploration to each other, and to the human operator as possible. Furthermore, human operators can efficiently control the robot team by specifying their interests in terms of these learned scene descriptors. The automatically generated exploration trajectories aim to maximize the information content, human interest, and spatial coverage, while taking into account the difficult constraints imposed by communication range in such harsh environments.
PIs: Yogesh Girdhar, Brian Claus
This work is supported by the National Science Foundation under Grant No. 1734400
