PhD Proposal Review: SpatioTemporal Prediction of Object Handover for Human Robot Collaboration
Sadjad Asghari Esfeden
Location: Zoom Link
Abstract: Predicting human behaviour in video is one of the challenging problems in computer vision. In order for robots to be able to interact with humans they need to understand human intent. We study the problem of object handover, where a robot tries to follow its collaborator’s movement as well as the object of interest to grasp the object in a human-like behavior. Therefore, the robot should predict a moving object’s time and location of handover. We propose a computer vision based algorithm to help robot understand its environment, detect, track, and predict object and human motions during the task of handover. The perception system enables robot to move towards the locus of handover before it occurs, and refine its motion when there is a change in human intention.