Trajectory Augmentation for Robust Neural Locomotion Controllers

Dhruv Agrawal, Dhruv Agrawal, Mathias König, Jakob Buhmann, Robert Sumner, Robert Sumner, Martin Guay

2024

Abstract

Neural Locomotion Controllers are promising real-time character controllers that can learn directly from motion data. However, the current state of the art models suffer from artifacts such as Pose Blocking and Foot Skating caused by poor generalization to real world control signals. We show that this is due to training on unbalanced biased datasets with poor representation for many important gaits and transitions. To solve this poor data problem, we introduce Trajectory Augmentation (TrajAug), a fully automatic data augmentation technique that generates synthetic motion data by using motion matching to stitch sequences from the original dataset to follow random trajectories. By uniformly sampling these trajectories, we can rebalance the dataset and introduce sharper turns that are commonly used in-game but are hard to capture. TrajAug can be easily integrated into the training of existing neural locomotion controllers without the need for adaptation. We show that TrajAug produces better results than training on only the original dataset or a manually augmented dataset.

Paper Citation

in Harvard Style

Agrawal D., König M., Buhmann J., Sumner R. and Guay M. (2024). Trajectory Augmentation for Robust Neural Locomotion Controllers. In Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 1: GRAPP; ISBN 978-989-758-679-8, SciTePress, pages 25-33. DOI: 10.5220/0012273300003660

in Bibtex Style

@conference{grapp24,
author={Dhruv Agrawal and Mathias König and Jakob Buhmann and Robert Sumner and Martin Guay},
title={Trajectory Augmentation for Robust Neural Locomotion Controllers},
booktitle={Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 1: GRAPP},
year={2024},
pages={25-33},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012273300003660},
isbn={978-989-758-679-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 1: GRAPP
TI - Trajectory Augmentation for Robust Neural Locomotion Controllers
SN - 978-989-758-679-8
AU - Agrawal D.
AU - König M.
AU - Buhmann J.
AU - Sumner R.
AU - Guay M.
PY - 2024
SP - 25
EP - 33
DO - 10.5220/0012273300003660
PB - SciTePress