Control of Physical Human-Robot Interaction: Mimicking Human Assistance

Engineering an assistive robot, capable of serving human needs and performing daily chores, has been a long-sought-for goal for the Robotics field as a whole. One of the main challenges facing researchers is on how to build the robot to be accepted by humans. There are many factors involved in having a robot and a human effectively collaborating, including technological limitations, anthropomorphic elements, ethical concerns, social factors, etc. One of the less explored aspects of this problem is physical interaction between a human and a robot. Envision a robotic assistant that is helping a human, moving a piece of furniture. Since the human and the robot are haptically coupled, every small movement/force of the robot is perceived by the human and can be interpreted as a clue for the next action. At the same time, the human expects the robot to understand the cues he/she is giving. In other words, the human expects the interaction to be fluid and natural, as it is with a human partner. Note that in a physical interaction between two humans, the kinesthetic cues serve as a communication channel that guarantees the success of the collaboration, even in cases when the verbal communication is missing. In this thesis, we focus on the physical interaction between a human and a robot. We first study the characteristics of a natural human-human physical interaction and explore different features of cooperation between two humans. In particular, we propose an abstract model for the quality of cooperation, a mathematical model for the motion trajectory during the interaction and a novel approach in modeling the interaction force between two humans. Based on these models that we construct for a natural human-human interaction, we propose a set of control policies that replicates the same interaction features and mimics human’s behavior during a physical interaction between a human and a robot.