Probabilistic Generative Modeling for Procedural Roundabout Generation for Developing Countries

Zarif Ikram¹, Ling Pan², Dianbo Liu^2,3

¹Bangladesh University of Engineering and Technology ²MILA ³NUS
NeurIPS 2023 Workshop on Adaptive Experimental Design and Active Learning in the Real World

Paper Code arXiv

Abstract

Due to limited resources and fast economic growth, designing optimal transportation road networks with traffic simulation and validation in a cost-effective manner is vital for developing countries, where extensive manual testing is expensive and often infeasible. Current rule-based road design generators lack diversity, a key feature for design robustness. Generative Flow Networks (GFlowNets) learn stochastic policies to sample from an unnormalized reward distribution, thus generating high-quality solutions while preserving their diversity. In this work, we formulate the problem of linking incident roads to the circular junction of a roundabout by a Markov decision process, and we leverage GFlowNets as the Junction-Art road generator. We compare our method with related methods and our empirical results show that our method achieves better diversity while preserving a high validity score.

Given incident road configurations, we generate a roundabout by inferring the circular structure from the incident roads. A valid solution is a roundabout that connects all the incident roads with any overlap between the connecting roads.

MDP design for incident road configuration generation. Number on dark filled circles on the dotted circle represent the endpoints of circle road segments. Small filled circle with arrow represent the road definition where direction of the arrow represent the heading of the initial road.

To reduce time complexity of validating a roundabout after construction, we propose a proxy reward that estimates the validation using straight line intersection. Left: Geometry of a generated roundabout. Right: Geometry generated by the proxy reward function.

Ablation study on GFlowNet training. We see the number of modes sampled increases with hidden dim and base value.

Poster

BibTeX

@inproceedings{
        ikram2023probabilistic,
        title={Probabilistic Generative Modeling for Procedural Roundabout Generation for Developing Countries},
        author={Zarif Ikram and Ling Pan and Dianbo Liu},
        booktitle={NeurIPS 2023 Workshop on Adaptive Experimental Design and Active Learning in the Real World},
        year={2023},
        url={https://openreview.net/forum?id=WWqJWiyQ2D}
        }