Underwater Mobile Manipulation (UMM)

The Underwater Mobile Manipulation project will develop and demonstrate key elements of an underwater robotics system that can provide unprecedented levels of perception and manipulation that will meet the challenges of the offshore energy sector. This objective will be met by developing an advanced perception suite for underwater manipulation and adapting the dexterous robotic manipulation technology suite demonstrated on the DARPA Robotics Challenge (DRC) RoboSimian robot for underwater applications.

The Underwater Mobile Manipulation project will perform technology development in the areas of perception, robotic limb hardware and robotic limb control in order to develop a subsea automated manipulation system. This development will build on capabilities demonstrated on other JPL limbed robotic platforms such as the Curiosity Rover, ATHLETE, and RoboSimian. The Underwater Mobile Manipulation robotic system will consist of a perception suite, 7 degree of freedom robotic limb and associated perception and control software. The perception suite includes color stereo cameras in dome-port underwater housings, a 2D imaging sonar, a variable illumination underwater light and a pan-tilt mount. Fused opti-acoustic point clouds will be created from these instruments using JPL stereo vision and sensor fusion techniques and acoustic image processing techniques described in literature. The robotic limb will be adapted from the modular limb design used by RoboSimian. Each joint will employ the same high power-density harmonic geared actuator. Actuator control electronics will reside in the actuator housings. The limb will incorporate a 3 DOF gripper based off of the RoboSimian CAMHand. On both the limb actuators and gripper, traditional o-ring and spring-energized sealing approaches will be used to seal the air-filled housings to 30m depth. To control the limb we will adapt the high DOF motion planning from ATHLETE and RoboSimian which make use of distributed, networked motor control components. Semi-autonomous manipulation behaviors will be developed by parameterizing behaviors to chain together sets of arm motions while monitoring end-effector force measurements. Scene reconstruction products, such as targets or collision volumes, will act as inputs into behaviors.