This thesis investigates a reactive motion planning and controller framework that enables robots to manipulate objects dexterously. We develop a robotic platform that can quickly and reliably replan actions based on sensed information. Robotic manipulation is subject to noise due to uncertainty in frictional contact information, and reactivity is key for robustness. The planning framework has been designed with generality in mind and naturally extends to a variety of robotic tasks, manipulators and sensors. This design is validated experimentally on an ABB IRB 14000 dual-arm industrial collaborative robot.
In this research, we are interested in dexterous robot manipulation, where the key technology is to move an object from an initial location to a desired configuration. The robot makes use of a high resolution tactile sensor to monitor the progress of the task and drive the reactive behavior of the robot to counter mistakes or unaccounted environment conditions. The motion planning framework is integrated with a task planner that dictates the high-level manipulation behavior of the robot, as well as a low-level controller, that adapts robot motions based on measured tactile signa
