Learning human-robot handovers through π-STAM: Policy improvement with spatio-temporal affordance maps | IEEE Conference Publication | IEEE Xplore