Real Time Motion Control For Natural Human Robot Interaction

Abstract

In this thesis, we present the research work performed with various robots including Android PKD, Zeno and the Atlas robot. The main focus of this thesis is to develop algorithms for vision based natural human robot interaction in real time. Zeno is a childlike robot that we use for the research related to Autism in children. Children with Autism who lack motor skills tend to shy away from humans. In this project we use the robot Zeno that is capable of facial expressions, arm and body motions to teach motor skills to them. For this purpose we developed a program that plays predetermined gestures like hand wave, tummy rub and fist bump in a particular sequence. This was used in the UNT Health science center to record the motion of the Autistic children so that we can compare the motion of the child and the robot. Android PKD is an avatar of the science fiction writer Philip K. Dick. In this project we developed a novel algorithm to distribute the motion between the neck and the eye servo motors to solve the redundancy involved. The robot was able to track faces by moving its eyes and neck in a realistic humanlike motion. Another project was done to mimic the facial expressions and the 3D rotations of the human head by the PKD. In this the mapping between the facial feature space and the actuator space is addressed using linear and nonlinear methods. Both of these demos and Zeno were exhibited at the Humanoids 2013 conference in Atlanta. An algorithm to map the human arm, torso and leg motion to the robot was developed as a project for DARPA Robotics challenge and also controlling the robot in real time. The robot used is called Atlas and it is a virtual robot in a Gazebo simulation environment. The issue of singularities has been addressed while mapping the human motion to the robot.