Reward hacking playground

Set a delivery robot’s reward. Watch the best-scoring behaviour diverge from the job you intended.

How this experiment works

Each strategy has deliveries, checkpoint visits and energy cost. Your coefficients determine its score. The chosen strategy maximises that explicit reward.

What it does not prove

This is an exhaustive comparison of three toy strategies, not reinforcement learning training. Its small world makes a general problem visible.

Try this challenge

Make successful deliveries the best choice, while still penalising wasted energy.

Enable JavaScript to run the experiment on your device.