Machine learning agents
Unity has released an open beta version of its machine learning (ML) agents. The software development kit is designed to help developers and researchers transform games and simulations into environments where intelligent agents can be trained. Unity relies on common learning methods such as deep reinforcement learning or evolutionary strategies, and a Python API. On its website the game engine developer introduces some applications of its ML agents, for instance stating that within 30 seconds, a single neural network is able to analyse the data of 12 agents to learn how to balance a ball on a floating platform. In order realise the full potential of the technology however, Unity is asking the development community for help. Basic algorithms and the code versioning of the beta release are available on GitHub.
We’d like to thank everyone involved at Unity for sharing the production status of their ML agents!
Read the full article here