Controlling character via high-level control signal sequences is of high desire for game designers and storytelling tasks. This work continues the recent work on MANN (Mode-Adaptive Neural Networks) for character control by the ability to carry out desired motion action types at a specified time or position using an interactive user-system. The system allows an offline control for different quadruped character movements, such as locomotion and stylizations thereof, sneaking, eating, and hydrating has been created. Additionally, this work proposes a new dataset which aims to enhance synthetic motions when trained jointly with motion capture data. The approach is based on manipulating postures using inverse kinematics.
- Video - Dataset - Windows Demo - Linux Demo - Mac Demo - ReadMe -
The motion capture data is available only under the terms of the Attribution-NonCommercial 4.0 International(CC BY-NC 4.0) license.