/reinforcement-learning-an-introduction--exercises

exercices from 'reinforcement learning an introduction by Sutton and Barto'

Primary LanguagePython

reinforcement-learning-an-introduction--exercises

Dynamic programming: car rental problem

Although dynamic programming is not reinforcement learning, it bears a lot of resemblence. Still it requires a full model of the environment and a lot of computation.

Dynamic programming was used here to solve jack's car rental problem as described in 'reinforcement learning an introduction' by Sutton and Barto.

text

Cartpole balancing

Open AI gym was used as simulation environment for balancing a cartpole.

cartpole simulation

Using reinforcement learning (Q-learning, off-policy) with a tabular approach the following control scheme was achieved after 100000 episodes of training.

balancing cartpole