Combining Static and Dynamic Predictions of Transfer Points for Human Initiated Handovers

Janneke Simmering, Sebastian Meyer zu Borgsen, Sven Wachsmuth, Ayoub Al-Hamadi

Abstract: In many scenarios where robots could assist humans, handover situations are essential. But they are still challenging for robots, especially if these are initiated by the human interaction partner. Human-human handover studies report average reaction times of 0.4 s, which is only achievable for robots, if they are able to predict the object transfer point (OTP) sufficiently early and then adapt to the human movement. In this paper, we propose a hand tracking system that can be used in the context of human initiated handover as a basis for human reaching motion prediction. The OTP prediction implemented is based on the minimum jerk model and combines a static estimation utilizing the human’s initial pose and a dynamic estimation from the current hand trajectory. Results are generated and analyzed for a broad spectrum of human initiated scenarios. For these cases we examine the dynamics of different variants of the proposed prediction algorithm, i.e., how early is a robot’s prediction of the OTP within a certain error range? The tracking delivers results with an average delay, after the initialization, of 0.07 s. We show that the OTP prediction delivers results after 75 % of the movement within a 10 cm precision box.