Adaptation of task-aware, communicative variance for motion control in social humanoid robotic applications

In this thesis, I explore the interaction benefits gained when robots communicate with their partners using a familiar mode: robot motion that is human-like. To achieve this, I have two concrete goals: (1) synthesize robot motion that is more human-like, and (2) show that communicative robot motion has benefits for interaction with human partners. These two goals are motivated by the limitations with existing motion generation methods in robotics, such as heavy data dependence, and the problems caused by the use of these existing techniques for human-robot interaction, such as long wait times, ambiguous motions, and unsynchronized collaboration.;Unfortunately, the majority of existing motion generation techniques for social robots do not produce human-like motion. For example, retargeting human motion capture data to robots does not produce human-like motion for robots because the degrees-of-freedom differ in number or location on the kinematic structures of robots and humans. Also, in the rare instances when retargeting human motion to robots works well, it produces only one motion trajectory, rather than a variety of trajectories, which makes the robot move in a very repetitive way.;After presenting algorithms for the synthesis of three methods of communication in motion, I develop two algorithms that focus on explicitly making robot motion more human-like. When motion is more human-like, the motion is more familiar to human partners, and they can more easily identify the robot motion correctly. In order to measure success, I present and validate a metric that is used to measure human-likeness of motion, so that robot trajectories can be evaluated quantitatively, allowing for simple comparison of motion on the same robot or between multiple robots with different kinematics or dynamics.;When synthesizing communicative motion or modifying trajectories to appear more human-like, the tasks the robot must accomplish in the world must not be disrupted and real-world interaction should not be hindered (i.e. motion should be task-aware). In the literature, motion control is comprised of joint-coordination and planning [5]. Joint-coordination involves identifying and implementing programs to move, which is what my two prior goals (i.e. communication and human-like synthesis) accomplish. On the other hand, planning develops algorithms for combining motions for autonomous interaction with a dynamic environment, such as obstacle avoidance and manipulation . Thus, the last sections of my thesis are devoted to integration of the algorithms (i.e. composing motions) and handling task-based constraints so the social robot can interact with the real-world.;My thesis discusses the results of 31 experiments, balanced between quantitative analysis and user studies, cumulatively involving 1,624 participants, and over 203,000 data points to support the interaction benefits of using my comprehensive algorithm to create task-aware, communicative, human-like motion for social robots. This large quantity of data is driven by the large number of experimental conditions per study and the expected variance of subjective opinions being large. Both these criteria lead to the requirement of many participants to achieve significant differences.;Discussion of all the results is distributed throughout the thesis, near the appropriate sections of the algorithm, but for example, my thesis concludes that the addition of three specific methods of communicating via motion (i.e. secondary motion, exaggeration, and anticipation) allow human partners to more quickly and more accurately tell what the robot is doing, feel more engaged in the interaction, and remember the interaction more accurately. Furthermore, communicative motion is shown to be more human-like and can be harnessed to direct the attention of the human partner. (Abstract shortened by UMI.).