This repo contains code to accompany the CORL 2020 paper: Explicitly Encouraging Low Fractional Dimensional Trajectories Via Reinforcement Learning. The root directory contains the original manuscript as a pdf, a short video, and the source code to replicate the results found in the paper.
The source code most notably contains a modified implementation of Augmented Random Search which modifies rewards obtained from reinforcement learning environments in order to explicitly encourage agents to find policies which induce trajectories with a small fractional dimension. There are also notebooks which analyze the resulting policies.