Abstract
As the percentage of the elderly population is rising worldwide, the demand imposed on modern society for providing a sufficient amount of skilled workers in the caregiving sector is becoming increasingly harder to fulfill. Utilization of robotic technologies in physical assistance in a home support context can contribute to sustaining a frail person's autonomy and quality of life. We envision the use of the humanoid robot technology for providing daily assistance with physical motion tasks. More specifically, in this work, we focus on the usage of Pepper humanoid robot platform, mass-produced by SoftBank Robotics. The choice of the platform is motivated by its affordability, user-friendly design and multi-modal communication capabilities.First, we propose and develop a proprioceptive sensor based contact detection method. In order to maintain low cost of the platform, our method for contact detection aims at using only the available Pepper’s sensors to detect collision with the environment, namely a contact event during physical human-robot interaction. We detail the integration of the proposed method as a feedback signal in the whole-body controller to react to human touch in real-time.Secondly, we investigate the whole-body humanoid robot posture planning problem in the assistive physical human-robot interaction context. We augment the non-linear optimization based posture generation framework with necessary components that allow us to plan a robot attitude in contact with a human point cloud. The proposed human point cloud processing pipeline provides the necessary data structures to formulate the posture generation problem for a robot to initiate a physical assistance task.Then, we present a fully autonomous interaction scenario for initiating a physical assistance process. A Finite-State Machine and task-space Quadratic Programming based controller is developed for a robot to navigate towards a human, perform a multi-modal communication and establish several physical contacts in a fully autonomous fashion. The controller performance is demonstrated in real experiments with a human subject. All the software tools developed to perform whole-body task space Quadratic Programming based control on SoftBank humanoid robots are made publicly available and are documented in detail.Finally, we study the problem of partial physical assistance in motion. We present a control methodology that enables a humanoid robot to supply the assistive forces necessary to help a frail human to achieve a desired performance of a motion task. We present and discuss the simulation results of the proposed method.We conclude this work with discussion of the achieved results and the future perspectives of research in the area of humanoid-human interaction for physical assistance.