Enduro Keeping-Pace Demonstration — inverts the scoring objective of → Enduro (Atari game)

The standard Enduro task and its contractor instructions define good play as passing as many cars as possible while not getting passed (deep-rl-human-prefs, §"B.2 Atari", p. 17). The keeping-pace demonstration retrains the identical game and action space under a literally inverted instruction: for "Enduro (even mode)" contractors are told to "Avoid passing cars OR getting passed by them," rewarding the agent for staying level with traffic instead of overtaking it (deep-rl-human-prefs, §"B.2 Atari", p. 17). Only the comparison instructions given to the human change; the environment, action space, and learning algorithm are the same as the benchmark task. Using roughly 1,300 queries and 4 million frames, the demonstration shows the preference-learning method can retarget a game's entire objective, including reversing it outright, without touching the game's code (deep-rl-human-prefs, §"3.2 Novel behaviors", p. 9).