Okay, so what do we have here?
In last years, novel frameworks for controlling and animating characters were presented, and have caused a lot of excitement in the animation community. In contrast to traditional techniques which rely on standard keyframe animation, the new systems were able to learn the animations from real motion data using deep neural network architectures. This Unity package now contains a ready-to-use implementation of such framework for handling your biped and quadruped locomotion.

What do you get?
With this package, you will get one of the best character controllers for humanoids and quadruped existing in the world. The characters are controlled via simple commands (i.e. movement-direction, target velocity and action), and the neural network then naturally performs proper transitions and motions. The package also comes with a post-processing solver to traverse over uneven terrain, stairs or small obstacles. This includes a fast and generic full-body IK solver that works independently from the neural network, and can be used via scripting for any other of your projects too.

What to know?
To achieve this quality, the system requires roughly 2-4ms computation time at each frame, and needs your game to run at either 30/60 FPS. Controllers for both framerates are provided. The neural network is bound to the skeleton, so the character mesh can be easily replaced after skinning it for the provided biped skeleton. Finally, the generated motion is subject to the control parameters. That means you can produce different details of motion by changing them in the inspector.

This asset is under active and ongoing research. Future progress will also include future updates for this package.

Make sure to run the demo first before purchasing! If any issues, please let me know. I am happy to resolve problems.

Supported Platforms
Windows / Demo
Mac / Demo
Linux / Demo