Goal: manipulate objects with flat surfaces underwater.
Solution: underwater vacuum-powered gripper with four-bar actuation.
Implementation: we developed a top-down object picking mechanism which uses a vacuum-powered suction gripper attached to a classic four-bar linkage. The linkage is actuated with an underwater servo motor while the vacuum is created by a brushless motor-powered marine thruster.
Goal: build maps of objects and obstacles in the environment
Solution: fusion of scanning sonar and RGBD video streams
Implementation: to develop environment maps and scene understanding, we employ a scanning sonar as well as stereo depth cameras. Combining classic computer vision approaches with deep learning, our perception stack is able to localize and track objects with visual features, like the buoys, while also incorporating objects beyond the visual range into the map using sonar.
Goal: localize underwater acoustic pingers
Solution: phased array of marine hydrophones
Implementation: to aid in mission and path planning, acoustic pingers guide the submarine towards tasks far from the starting location. To localize these pingers, we developed a phased array of four hydrophones capable of determining the direction of arrival (DoA) of incoming ultrasonic pings. The operating principle behind this system lies in the time differences of arrivals of pings across the block of receivers, where phase shifts between each sensor indicate the DoA of the ping.