Figure 4
Success rate of observer AI and Physical scenes and outcomes envisioned by the observer. (A) We report the successful rate of the observer AI for each type of behavior during testing. Noted that all the behaviors of the actor robot are seen during training together and no other information is given except the single image frame. Our observer AI achieves a 98.5% success rate on average across all four types of behaviors of the actor robot. (B) The first column and the third column show some sample scenarios involving the actor robot, one or two green circles, and a square obstacle. The second and the fourth column show the outcome as envisioned by the observer. The left two columns (Success) show successful vision, whereas the right two columns (Failure) show failed envision.
