Some experiments judge the end condition too early.
For example, in the stack cube task, it is marked as successful when the gripper merely hovers above the target. It should instead wait until the gripper opens and the cube is stably stacked without falling.
In the put eggplant task, it is marked as successful when the gripper touches the edge of the sink dish rack. It should instead wait until the gripper moves to the center of the rack.