Towards a software pipeline for seamless robot-to-human handover

This work presents a modular software architecture for robotic handover. The architecture includes: neuromorphic perception for detection and estimation of human movements; motion predictors for anticipating partners movements and adapt to them; and finally, adaptation mechanisms that command a low-level controller for a robotic arm equipped with an anthropomorphic hand, for object seamless motion and release. Preliminary results in robot-to-human handover involving the modules of prediction and low-level control corroborate the effectiveness of the proposed architecture.