Remote robotic manipulation task execution using affordance primitives

In recent years, an increasing number of robots have been deployed to an increasing number of environments to do an increasing number of tasks. Often the goal is for these robots to operate with as little human intervention as possible, and this remains a challenge for tasks that involve sustained physical interaction between the robot and environment. The problems compound if the robot is located far from its operators, where the remote connection between robot and operator may have significant latency, making both control and feedback difficult. Addressing these challenges by increasing autonomy for such contact tasks is critical for scaling robot deployment outside of lab or manufacturing settings, and into the real world. This dissertation focuses on improving the ability of robots to complete contact tasks with objects designed for humans, such as doors, valves, etc. These objects are inherently interesting to manipulate (i.e. because they do something), vary widely, and are widespread in many environments. Manipulating them effectively expands the areas robots may be deployed for manipulation, and has navigation implications as well (e.g. traversing doors or elevators). This work introduces affordance primitives as the instantaneous spatial and force profile for contact tasks, and uses them to improve robotic manipulation of articulated objects. Affordance primitives may be used as virtual fixtures during real-time teleoperated control, or autonomously with a flexible and reactive motion planner presented in this dissertation. They are based on screw theory, and naturally capture simply articulated objects with one or more degrees of freedom. The implementation of affordance primitives is agnostic to the task, environment, and robot hardware, and allows flexibility in the control input from the user or autonomously. The contributions of this work include reducing operator burden during teleoperated tasks, modeling generic tasks using affordance primitives, and enhancing autonomous execution of tasks modeled with affordance primitives. Using affordance primitives leads to a 5x improvement in task duration and 3x reduction in contact forces over prior work while being robust to errors in the task model up to 10cm and 20°. This dissertation advances the capabilities of robotic manipulators in remote settings, and works to address the robot scaling problem to deploy more manipulators to the real worldMechanical Engineerin