Benchmark results

[MeixinZhu]

Thanks for this great project!! Are there any benckmar results on which model + obs + reward work best in a specific environment? @eleurent

[eleurent]

Unfortunately not. In this paper, I benchmark three pairs of (obs, model), but that's about it. But that would be a great idea, if you plan on doing a more comprehensive benchmark with reproducible experiments, that would make a nice addition, so feel free to do a PR.

[MeixinZhu]

Thanks! Yes, I will make a PR if I have some results about this. Another question is when I directly use steer and acceleration as continuous action (the continuous action type), the performance is really bad (vehicles will just rotate). Any suggestions for using continuous action?

[MeixinZhu]

@eleurent

[eleurent]

Sorry for not replying, let me know if this is still needed.