RFM-1 update: In-context learning to improve grasping