Closed edbeeching closed 8 months ago
I'd just like to add on this topic, that due to an issue, action definitions may need to be sorted alphabetically for some of the envs to work properly with onnx inference, at least until it gets resolved: https://github.com/edbeeching/godot_rl_agents_plugin/issues/25
Some may need the plugin updated with the newest version from the plugin repository to fix some onnx inference errors. For now it's sufficient to copy the addons folder and csproj file contents from: https://github.com/edbeeching/godot_rl_agents_plugin/tree/main
Thanks, I will update the plugin to the latest version in all examples in this process.
Thanks for offering to tackle this. Let us know if you need any help with it.
I have just remembered that one of my envs is using a slightly modified plugin, so in case or AirHockey we shouldn't change the plugin at this point (the plugin used is recent enough that onnx inference should work as is).
For future reference, I changed this line: https://github.com/edbeeching/godot_rl_agents_examples/blob/bbe18a44eac8f5576f2b746c9f484ce4a488dfed/examples/AirHockey/addons/godot_rl_agents/controller/ai_controller_3d.gd#L50
This dependency can be removed at some point by adding the change to the AIController override instead as I did in my later env https://github.com/edbeeching/godot_rl_agents_examples/blob/bbe18a44eac8f5576f2b746c9f484ce4a488dfed/examples/3DCarParking/scenes/car/CarAIController.gd#L62
3DCarParking itself is new so it may already be using the newest plugin, but I think it should work fine if the plugin was replaced.
I could take care of this.