Check out Learning Agents, a new UE5 plug-in that allows you to train AIs using reinforcement and imitation learning.
In case you missed it, back in late March, Research Engineer at Epic Games Brendan Mulcahy shared the first look at Learning Agents, an impressive new plug-in for Unreal Engine 5 that allows users to train NPCs through reinforcement and imitation learning, greatly simplifying the development of game-playing agents, physics-based animations, automated QA bots, and more.
In the announcement, Mulcahy emphasized that the new plug-in makes it possible to enhance or replace traditional game AI, such as those implemented with behavior trees or state machines, with the long-term goal of being useful in a range of applications. The tool was designed first and foremost for game developers, with the primary intention of simplifying the training and deployment of machine learning bots in their existing or upcoming games.
Five months later, the plug-in has finally been released and made available to developers from all over the world with the launch of Unreal Engine 5.3 – a massive update that brought enhancements to the engine's existing tools (Lumen, Nanite, Path Tracing, etc.) and introduced tons of new experimental features for rendering, animation, and simulation.
If you would like to get started with Learning Agents in UE5.3, the best place to start would be Mulcahy's comprehensive written tutorial "Learning to Drive", which thoroughly explains how the novel framework can be leveraged to teach a neural network to drive a sports car around a race track.
With this tutorial, you'll learn how to train an agent that can efficiently complete the track without veering off course, utilizing reinforcement learning. The tutorial takes advantage of the pre-installed Vehicle Template in Unreal Engine 5.3, saving you valuable setup time. It's worth noting that this tutorial is designed for Blueprint usage only, so you won't need any prior knowledge of C++. You can access the full tutorial by clicking this link.
Speaking of machine learning, in UE5.3, alongside the introduction of the Learning Agents plug-in, there have been significant enhancements to the Machine Learning Deformer Sample. This sample is designed to illustrate how Unreal Engine's ML technology can be employed to create high-fidelity next-generation characters with deformations driven by full muscle, flesh, and cloth simulations.
The release introduces a new performance-boosting feature called the Bone Mask and includes the Houdini scene used by the development team for training. You can learn more about it here.