Situated robot learning for multi-modal instruction and imitation of grasping