Ml-agents: tf-agents

[mattinjersey]

I wonder if you have considered collaboration with tf-agents. they are also building a reinforcement library.

[roboserg]

Unity ML has Python API. With that, you can use whatever library you want. Also, see their gym interface, then you can copy paste any library / tutorial online.

[ervteng]

Thanks for the heads-up @mattinjersey! When we first looked at TF-Agents, it was in (very) early forms. But definitely worth another look now that it's semi-official. If anything, a shim between tf-agents and the ML-Agents Python API (or Gym wrapper) shouldn't be too hard.

We've also successfully integrated Baselines and Dopamine using the Gym wrapper.

[unityjeffrey]

thanks @mattinjersey and @roboserg - we've added this in our backlog to further explore. thanks for bringing it up