As someone working on a reinforcement learning/neuroevolution problem right now, I find this to be extremely exciting. Fewer parameters, ceteris paribus, is always better—the fact that the experiments in this paper were run on one workstation, rather than on a massive farm of TPUs à la AlphaGo, implies quicker development iteration time and more accessibility to the average researcher.
The staging of components in this paper (compressor/controller), where neuroevolution is only applied to a low-dimensional controller, reminds me of Ha and Schmidhuber's recent paper on world models (which is briefly cited) [1]. They employ a variational autoencoder with ~4.4M parameters, an RNN with ~1.7M parameters, and a final controller with just 1,088 parameters! Though it's recently been shown that neuroevolution can scale to millions of parameters [2], the technique of applying evolution to as few parameters as possible and supplementing with either autoencoders or vector quantization seems to be gaining traction. I hope to apply some of the ideas in this paper to multiple co-evolving agents...