This thesis addresses the problem of a robotic arm striking an object in free flight to a target, and a closely related problem of estimating its pose and motion during the flight based on vision. Both problems share a common theme of dealing with a free flying object that is often not considered in robotics applications. Our investigation draws upon techniques from robotic manipulation, planning, mechanics, state estimation, aerodynamics, and camera modeling.

The robot batting problem is a skillful task that requires accurate perception of the flying object, robust modeling of impact dynamics, and efficient planning of a robotic arm's motion. Leveraging of impact and measuring motion are of great importance in manufacturing, sports, and space robots. To demonstrate the use of impact, we solve the batting problem in two dimensions based on impulse and energetic restitution with friction, flight mechanics incorporating gravity and aerodynamic forces, and trajectory re-planning for the bat-wielding robotic arm. Experiments with different objects show better batting performance than a human with no training.

The component of estimating the pose and motion of an in-flight object is subsequently extended to three dimensions. We present a stereo vision system consisting of two high-speed cameras. A hypothesis-based algorithm is proposed to track the object's features across a sequence of images, and for each active hypothesis, Kalman filtering is employed to compete for the estimation of linear and angular motions. A constrained Kalman filter is introduced to handle multiple quadratic constraints from the estimation of quaternions. Aerodynamic forces for a general shape are modeled through computational fluid dynamics, and the constraint of two-view geometry from stereo cameras is considered in measurements from images. Results have been obtained from the flights of two objects, and compared against calculations based on accelerometer data and image coordinates.