Hello all,
I’m writing to announce the use of the BHumanCodeRelease (2023) version for the WisTex United team. We consider our codebase to be derived software as we use the BHuman motion control, localization, perception and communication modules. The specific contribution of our work (taken from our SPL 2025 application) is as follows:
Our team codebase for the 2025 competition is based upon the \href{https://github.com/bhuman/BHumanCodeRelease/releases/tag/coderelease2023%7D%... 2023 code release }. As described below, our primary contribution has been to replace the high level behaviors written by BHuman with new, learned behaviors trained with deep reinforcement learning. As discussed below, we believe these changes to be transformative in nature. Thus, as of now, our robots use the perception, state-estimation, low-level motion control and communication modules provided by BHuman. We thank BHuman for making their full software stack available. Additionally, we used the open source library \textbf{\href{https://github.com/DLR-RM/stable-baselines3%7D%7Bstable-baselines3%7D%7D to train our policies, using proximal policy optimization\cite{https://doi.org/10.48550/arxiv.1707.06347%7D%5Ccite%7Bstable-baselines3%7D.
We also note our additions this year are improvements in learned cooperation which substantially changes our code but generally leave the BHuman modules unchanged from their 2023 version.
We also will once again release our updated code before the end of the year.
Adam