This is the training environment for Reinforced Machine, a twin-stick shooter action game. Choose your favorite mech, each with their own unique skills. Unleash powerful attacks with the help of your AI-assisted weapons to aid you in your battle against legions of cyborg creatures.

The trained AI Agent models are used to customize the weapon's behavior in the game. Players select a weapon, then choose their preferred AI Agent model from a list of pre-trained models, community trained models shared publicy, or their own custom trained models.

There are two ways to train and run the Reinforced Machine training environment.

1. (Recommended) In Google Colab where all the needed dependencies will be installed and pull the latest Reinforced Machine Training Linux build.

   • https://colab.research.google.com/drive/1Y9a7H1_fKiyBObvVmjztJ9WC_Id7qhb1

2. Train it locally. Here you need to setup Unity ML Agents Release 21 and install Pytorch along with all its dependencies.

   If you wish to train locally, follow the steps on how to install Unity ML-Agents in your local machine on this page.

   • Installing Unity MLAgents in Local machine

The next section is a guide on how to train the AI Agent. This applies to both training environments running in Google colab or in your local machine.

TODO: Create builds for Windows and Linux

Run this command to train with the Reinforced Machine executable build.

mlagents-learn ./config/ppo/RM_PPO_EnergyGun.yaml --env="./builds/RM_Game_Training_Only" --run-id=RM_PPO_EnergyGun_01 --time-scale=1

This sets the time-scale to 1. Remove it to make the AI Agent train faster.

After training the AI Agent, use the resulting ONNX model to run the build in Inference Mode.

TODO: Pass env variable with full file path of the ONNX model

Your trained AI Agent model can be used in the game. More details to follow near the release of the game.

Each weapon has its own implementation of the AI Agent. This means it has its own action space and some weapons have custom observations added to feed to the neural network.

Weapon	Behavior Name	Weapon ID	Training Document	Status
	EnergyGun	0	Energy Gun Training Document	Complete
	RocketLauncher	1	Rocket Launcher Training Document	TBD
	LaserGun	2	Laser Gun Training Document	TBD
	PortalGun	3	Portal Gun Training Document	TBD
	RailGun	4	Rail Gun Training Document	TBD
	AcidGun	5	Acid Gun Training Document	TBD
	SniperRifle	6	SniperRifle Training Document	TBD
	AssaultRifle	7	AssaultRifle Training Document	TBD
	ElectricGun	8	Electric Gun Training Document	TBD
	MachineGun	9	Machine Gun Training Document	TBD
	ShotGun	10	Shot Gun Training Document	TBD