Results for learning from multiple videos

In the table below, each row presents an:

original action id from the Something-Something dataset
one example video from that action
successful action by policy on simulator
failed action by policy on simulator
real robot performing action, video at 5x speed (note camera is from opposite direction)

Action	Example Video Demonstration	Simulator Success	Simulator Failure	Real Robot
86: Pull left to right				SUCCESS
87: Pull right to left				SUCCESS
93: Push left to right				SUCCESS
94: Push right to left				SUCCESS
47: Pick up				SUCCESS
104: Put behind				SUCCESS
105: Put in front of				SUCCESS
107: Put next to				SUCCESS
112: Put onto				FAILURE