AMaze

AMaze is a pure Python library for sighted learning agents. It combines low computational costs with a high capacity for customization to generate challenging, yet quickly computed, tasks. Its main use is as a benchmark generator to try out different implementation of algorithms from Reinforcement Learning to NeuroEvolution.

While the implementation is mostly done with a soon-to-be mature API, the characteristics and most favorable uses of the library are not yet fully explored. A few projects are currently underway to make progress in that direction:

• Impact of human agent in Interactive Reinforcement Learning
• CNNs for maze-navigating agents

In addition, open projects remain:

• Continuous control in maze-navigating agents
• Integration with existing libraries:
  • Stable baselines 3
  • ABrain
  • Anything else?

Feel free to contact me to participate.