Think of a game like dark souls, is it possible using ml-agents (using RL algorithms) to train monsters , to be able to compete against humans and make the game more interesting for players?
We have a self-play feature that might fit this need. The basic idea would be that the monster and player are agents that learn by competing against each other. Then after training, you can control the player yourself using a heuristic or player controller.
You can read more about self-play here ml-agents/docs/Learning-Environment-Design-Agents.md at release_16_branch · Unity-Technologies/ml-agents · GitHub. Also, please look at our Soccer and StrikersVsGoalie example environments for examples of using self play in symmetric and asymmetric games.
1 Like
Thank you @andrewcoh_unity , that is a good starting point
@andrewcoh_unity it is bit confusing for me, how self-play is used when the player has skill set (actions) very different from monster’s skill set (actions)