Extended Data Table 3.

Rewards used in the experiments

Empty cells are not used in that reward. Any cell that does not specify a weight has an implicit weight of 1. Vector-valued weights (for example, Droplets: R) return several values to the final combiner. See Exended Data Table 4 for the descriptions of the different reward components and Extended Data Table 5 for the transforms, combiners and terminations. All of the terminations criteria were used for these experiments. Code for these rewards is available in the supplementary material.