Reinforcement learning algorithms for the Untangling of Braids

Abstract

We use reinforcement learning algorithms (Q-Learning and Deep Q-Learning) to tackle the problem of untangling braids
and to compare the results of both algorithms. The idea is to use multi-agent (two competing players) based approach
to tackle the problem of untangling braids. We interface the braid untangling problem with the OpenAI Gym envi-
ronment, a widely used way of connecting agents to reinforcement learning problems. The results provide evidence
that the more we train the system, the better the untangling player gets for both approaches at untangling braids. The
comparison of both approaches produces interesting results, where Q- learning performs better while dealing with braids
of shorter length, whereas DQN performs slightly better while dealing with braids of longer lengt