Modeling Conditional Dependencies in Multiagent Trajectories

Yannick Rudolph, Ulf Brefeld
Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, PMLR 151:10518-10533, 2022.

Abstract

We study modeling joint densities over sets of random variables (next-step movements of multiple agents) which are conditioned on aligned observations (past trajectories). For this setting, we propose an autoregressive approach to model intra-timestep dependencies, where distributions over joint movements are represented by autoregressive factorizations. In our approach, factors are randomly ordered and estimated with a graph neural network to account for permutation equivariance, while a recurrent neural network encodes past trajectories. We further propose a conditional two-stream attention mechanism, to allow for efficient training of random factorizations. We experiment on trajectory data from professional soccer matches and find that we model low frequency trajectories better than variational approaches.

Cite this Paper

BibTeX
@InProceedings{pmlr-v151-rudolph22a, title = { Modeling Conditional Dependencies in Multiagent Trajectories }, author = {Rudolph, Yannick and Brefeld, Ulf}, booktitle = {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics}, pages = {10518--10533}, year = {2022}, editor = {Camps-Valls, Gustau and Ruiz, Francisco J. R. and Valera, Isabel}, volume = {151}, series = {Proceedings of Machine Learning Research}, month = {28--30 Mar}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v151/rudolph22a/rudolph22a.pdf}, url = {https://proceedings.mlr.press/v151/rudolph22a.html}, abstract = { We study modeling joint densities over sets of random variables (next-step movements of multiple agents) which are conditioned on aligned observations (past trajectories). For this setting, we propose an autoregressive approach to model intra-timestep dependencies, where distributions over joint movements are represented by autoregressive factorizations. In our approach, factors are randomly ordered and estimated with a graph neural network to account for permutation equivariance, while a recurrent neural network encodes past trajectories. We further propose a conditional two-stream attention mechanism, to allow for efficient training of random factorizations. We experiment on trajectory data from professional soccer matches and find that we model low frequency trajectories better than variational approaches. } }
Endnote
%0 Conference Paper %T Modeling Conditional Dependencies in Multiagent Trajectories %A Yannick Rudolph %A Ulf Brefeld %B Proceedings of The 25th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2022 %E Gustau Camps-Valls %E Francisco J. R. Ruiz %E Isabel Valera %F pmlr-v151-rudolph22a %I PMLR %P 10518--10533 %U https://proceedings.mlr.press/v151/rudolph22a.html %V 151 %X We study modeling joint densities over sets of random variables (next-step movements of multiple agents) which are conditioned on aligned observations (past trajectories). For this setting, we propose an autoregressive approach to model intra-timestep dependencies, where distributions over joint movements are represented by autoregressive factorizations. In our approach, factors are randomly ordered and estimated with a graph neural network to account for permutation equivariance, while a recurrent neural network encodes past trajectories. We further propose a conditional two-stream attention mechanism, to allow for efficient training of random factorizations. We experiment on trajectory data from professional soccer matches and find that we model low frequency trajectories better than variational approaches.
APA
Rudolph, Y. & Brefeld, U.. (2022). Modeling Conditional Dependencies in Multiagent Trajectories . Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 151:10518-10533 Available from https://proceedings.mlr.press/v151/rudolph22a.html.

Related Material