Elliptical Handover Reaching Motions
Project lead
Chief investigator
External collaborators
Sara Sheikholeslami (UBC)
Research topic Human-Robot Interaction, Modelling and Control
Industry application Assistive Robotics, Service robots
The service robotics sector has been rapidly growing in the recent decades, with much development for different applications such as homecare, hospital, manufacturing, and agriculture. In many of these service robot applications, object handover is a fundamental task that will frequently arise. Thus, efficient performance of handovers is essential to the effectiveness of robots belonging to this category.
While humans typically perform handovers with ease in various situations, it is still challenging for current robotic systems. Careful coordination between the two agents is required during the handover task to ensure successful transfer of the object from the giver to the receiver. From a sensing, coordination, and safety perspective, it is much more challenging than solo object manipulation tasks such as grasping and pick-and-place, where there is no human partner involved. There have been many works in the robotic literature addressing this problem, one common conclusion is enabling robots to employ humanlike behaviours improves human-robot handovers. During handover task, humans also utilize different communication channels (e.g., gaze, reaching motion, haptics, pose) during different phases (i.e., approach phase, reaching phase, object transfer phase, retracting phase) to coordinate the different aspects of handovers (i.e., where, when, how).
Among different communication modalities, this project focuses on generating humanlike reaching motions for robot. Inspired by earlier work from our collaborator from University of British Columbia on reaching motions, we first carried out an experimental validation of the same model and compare it with other established models using an existing dataset on unconstrained human-human handover (where the person is allowed to move their whole body). Finally, we implement a trajectory generator for the best models on a robotic arm with the aim of producing a more legible and predictable action to humans.