Perception and Control Challenges for Effective Human-Robot Handoffs

摘要

Human-robot handoffs are a primary task for a personal robot. Previously, robots put the majority of the handoff burden on the human, where in the case of the human to robot handoff the human had to position and orient the object into the robot’s hand. This works well because it is a very easy task for humans, but what happens if humans handoff objects while they are distracted or if they have some impairment? Is the handoff still possible? The recent arrival of reliable human-tracking systems like the Kinect and smart control algorithms that use the human-tracking feedback and other sensor data can allow for rich human-robot interaction where the robot is aware of the human and can actively collaborate together towards a final goal. In this paper, we use rich 3D RGB-D data for humantracking and object detection, combined with incremental control methods, to create a handoff system that is reliable and allows the robot to actively collaborate in a handoff with a human.