Results

Size Reduction - Locomotion

Size Reduction - Manipulation

For each task we see considerable decrease in network size required to achieve at least 90% performance.

Reward vs Capacity

For each task reward seems to be correlated with the capacity metric for each MLP.