This thesis explores an instance of Moravec's paradox within the domain of impact-aware non-prehensile manipulation. The specific task investigated here involves interacting with a simple environment through the application of an impulse, or a hit / strike. The objective is to endow a fixed redundant robot with the capability to hit an object within its workspace and project it to a desired location beyond its immediate reach.

In the initial part of this thesis, we deconstruct the interaction between the robot and the object through hitting into three key components: robot motion, contact, and the subsequent behavior of the object. This utilizes dynamical systems to generate the hitting motion for the robot, which allows for passive control and re-planning, i.e., disturbance robust motion. The robot strikes a known object at varying speeds and angles, and the resulting motion data is recorded. Gaussian Mixture Models are employed to learn the distribution of hitting speed, angle, and the distance the object moves. This forward model is then inverted to predict the necessary hitting speed to achieve a desired final position of the object.

The second part of this thesis tackles understanding the robot motion for impact. We introduce a metric termed "hitting flux", which relates the post-impact speed of the object with the robot's velocity, directional inertia, and the object's mass. To control the object's post impact velocity, one needs to control both the directional inertia and the velocity of the end effector. The robot's velocity is regulated through a passive velocity tracker, enabling it to respond appropriately to the impact. The robot's redundancy is leveraged to adjust its configuration towards the desired inertia value or maximizing the directional inertia value while leading to the same post-impact object behavior. Further, the study explores controlling the translational inertia of the robot's end effector within the task space to achieve the desired impact location.

The third part of this thesis examines how a redundant robot can generate impacts with joints other than its end-effector. Our observations indicate that to hit a heavier object and reduce the robot's rebound velocity, it is advantageous to increase the robot's inertia at the point of impact. Notably, the robot effective inertia is greater at joints other than the end effector. Although hitting speed decreases as the impacting joint moves inward, certain joints can still achieve the desired hitting flux. We propose a method for determining which joint to utilize for creating an impact and how to control the robot's motion accordingly.

The final part of this thesis develops a semi-autonomous dual-arm framework inspired by the game of air hockey to collect data on object motion resulting from an impulse. This framework enhances the object motion model established in the first part, making it largely independent of the hitting robot, its configuration, and properties of the object being hit. The system's efficiency is improved through the use of an Impact-Aware Extended Kalman Filter, which rapidly converges to an estimate of the coefficient of restitution and friction, thereby predicting the object's final position following an impulse. The collected data is utilized to model object motion demonstrating its applicability in enabling robot collaboration, where two robots cooperate to place an object at a distant location through successive hits, like in golf.