Reinforcement Learning with Neuroevolution

The assignment for this homework is short and simple: choose a few (2-3) environments from the gymnasium library and use neuro-evolution to solve them.

We used the library in the second practicals, your task is basically to integrate RL and neuro-evolution and test it. I recommend choosing some of the simple classic control environments, but if you want more challenge you can also try the Box2D or MuJoCo ones (e.g. LunarLander, Hopper, and Swimmer should not be that hard).

You can try NEAT, or you can try just the evolution of weights for a fixed topology, it is up to you.

For the solution, please, send me an e-mail that contains the list of problems you tested and the description of the algorithm you used together with its settings. Please, also include a plot that shows how the performance of the best individual changed during the evolution (e.g. as the reward achieved by the best individual).


The deadline for the assignment is May 4, 2023.