Breakout Variants from Schema Networks Paper

[jjfiv]

Schema Networks: Zero-shot Transfer with a Generative Causal Model of Intuitive Physics

• [x] Middle wall (unbreakable) "A wall is placed in the middle of the screen, requiring the agent to aim around it to hit the bricks."
• [x] half negative bricks
• [x] offset paddle: "The paddle is shifted upward by a few pixels."
• [x] random target - "A group of bricks is destoyed [sic] when the ball hits any of them and then reappears in a new random position, requiring the agent to delibarately [sic] aim at the group."
• [x] juggling, e.g., multiple balls - "Without any bricks, three balls are launched in such a way that a perfect policy could juggle them without dropping any."