Multisensory Integration For Adaptive Physical Human-robot Interaction

Abstract

Automated systems and have become increasingly prevalent in the 21st century. With their increased processing power, personal and home robotics has finally come within reach of the public. For robots to finally work with and around humans there are some challenges to be overcome. Some of these include complex social interaction, safety, identifying human intent, unpredictable physical interaction, and uncertain dynamic environments. This research focuses on safe, robust, and intuitive physical interaction with humans utilizing multiple sensor modalities.In this thesis a solution for adaptive physical human-robot interaction is proposed. The proposed framework consists of three parts. Dynamics compensation: adaptive dynamics compensation is proposed and extended to control complex, nonlinear, and constantly changing robotic systems. Adaptive force control: adaptive inverse control techniques are utilized to develop an admittance control system to compensate for changing task, sensor, and human conditions during interaction. Human motion study: Dynamic time warping is utilized as a method for movement characterization and dynamic movement primitives are used to develop a scalable and adaptable movement representation system.The proposed framework was inspired by the structure of the human motor cortex and somatosensory systems. An inner-loop control structure performs the function of the lower level fast dynamic compensation system while an outer-loop adaptive force controller enables task, sensor, and human specific control. This type of controller can adapt to the changing dynamics of the robot as well as compensate for the changing environmental sensing capacity and interaction scenario.