BumbleBee

Please wait a while for all the videos to load.

If you are unable to play the video, please try the alternative link: Alternative Link and refresh the page. Or, you can download the repo and watch it locally.

Let’s first provide a preview of the performance of our general policy. general

Long Sequences

We first demonstrate the performance on the LAFAN1 dataset, sourced from the Ubisoft La Forge Animation Dataset (Harvey et al., 2020) [1].

• Charleston Dance: Charleston Dance

• Boxing: Boxing

• Kung Fu: KungFu

• Merged motion combining Charleston Dance and Boxing: Merged

Motions in AMASS

We present the performance of the policies on the AMASS dataset. For motions that can be completed from the beginning, the left side shows their performance under the initial expert models, and the right side shows their performance under the final general policy. The remaining videos demonstrate the performance of the final policy.

Walk Fast

• motion 1: WF-motion1

• motion 2: WF-motion2

Jump

• motion 1: Jump-motion1

• motion 2: Jump-motion2

Stand Low

• motion 1: SL-motion1

• motion 2: SL-motion2

Stand Mid

• motion 1: SM-motion1

• motion 2: SM-motion2

Stand Up

• motion 1: SU-motion1

• motion 2: SU-motion2

Walk Slow

• motion 1: WS-motion1

• motion 2: WS-motion2

Reference

[1] Harvey, F. G., Yurick, M., Nowrouzezahrai, D., & Pal, C. (2020). Robust Motion In-Betweening. ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH), 39(4).