The following is an example of self-supervised robot learning (select all that apply):

Training a model to infer grasps from images, then using force/torque sensors on the wrist to automatically label executed grasps.

Training a model to infer the desired contact position and velocity of the end effector hitting a drum from images, then using resulting audio signals to label the strokes.

Training a model to infer a trajectory for the robot arm to place an object down, then time-reversing it to get a trajectory for picking up that same object.

Training a model to infer whether the vehicle should turn left or right given an image, then sending the image to the cloud to be labeled with crowd-compute.