Multi Agent Navigation in Unconstrained Environments using a Centralized Attention based Graphical Neural Network Controller

Yining Ma*1,2, Qadeer Khan*1,2, Daniel Cremers1,2,3
Technical University of Munich1, Munich Center for Machine Learning2, University of Oxford3
IEEE International Conference on Intelligent Transportation Systems ITSC 2023
*Indicates Equal Contribution
Paper Supplementary Code arXiv

!!! Our new work MA-DV2F with improved performance can be found here. !!!

Abstract

In this work, we propose a learning based neural model that provides both the longitudinal and lateral control commands to simultaneously navigate multiple vehicles. The goal is to ensure that each vehicle reaches a desired target state without colliding with any other vehicle or obstacle in an unconstrained environment. The model utilizes an attention based Graphical Neural Network paradigm that takes into consideration the state of all the surrounding vehicles to make an informed decision. This allows each vehicle to smoothly reach its destination while also evading collision with the other agents. The data and corresponding labels for training such a network is obtained using an optimization based procedure. Experimental results demonstrates that our model is powerful enough to generalize even to situations with more vehicles than in the training data. Our method also outperforms comparable graphical neural network architectures.


MY ALT TEXT

Our proposed U Attention Block:


MY ALT TEXT

Please refer to here for the detailed explanation of U Attention Block.



Results of our Model


Attention Mechanism of our Model

MY ALT TEXT

The GIF above shows how the attention changes when the vehicles are moving. We visualize the mean of attention logits from all the graph attention layers of our model. The rows in the attention matrix correspond to the vehicle of interest. The columns show which vehicles/obstacle is being attended to. A lighter shade in the attention matrix depicts high attention and a darker shade represents lack of attention.


Comparison with Other Models


BibTeX


      @INPROCEEDINGS{10422072,
        author={Ma, Yining and Khan, Qadeer and Cremers, Daniel},
        booktitle={2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC)}, 
        title={Multi Agent Navigation in Unconstrained Environments using a Centralized Attention based Graphical Neural Network Controller}, 
        year={2023},
        volume={},
        number={},
        pages={2893-2900},
        keywords={Training;Codes;Navigation;Neural networks;Training data;Data models;Optimization},
        doi={10.1109/ITSC57777.2023.10422072}}