Using Machine Learning to Enhance Vehicles Traffic in ATN (PRT) Systems

TL;DR

This paper explores how machine learning can optimize autonomous vehicle cooperation in Automated Transit Networks to improve throughput by tuning control algorithms and managing empty vehicles more effectively.

Contribution

It introduces ML-based refinement models to enhance existing ATN control algorithms, specifically improving empty vehicle management and system performance.

Findings

ML tuning improves vehicle reallocation efficiency

Enhanced algorithms increase system throughput

Behavioral policies tailored by ML outperform manual controls

Abstract

This paper discusses new techniques to enhance Automated Transit Networks (ATN, previously called Personal Rapid Transit - PRT) based on Artificial Intelligence tools. The main direction is improvement of the cooperation of autonomous modules that use negotiation protocols, following the IoT paradigm. One of the goals is to increase ATN system throughput by tuning up autonomous vehicles cooperation. Machine learning (ML) was used to improve algorithms designed by human programmers. We used "existing controls" corresponding to near-optimal solutions and built refinement models to more accurately relate a system's dynamics to its performance. A mechanism that mostly influences ATN performance is Empty Vehicle Management (EVM). The algorithms designed by human programmers was used: calls to empty vehicles for waiting passengers and balancing based on reallocation of empty vehicles to achieve better regularity of their settlement. In this paper we discuss how we can improve these algorithms (and tune them to current conditions) by using ML to tailor individual behavioral policies. Using ML techniques was possible because our algorithm is based on a set of parameters. A number of weights and thresholds could be tuned up to give better decisions on moving empty vehicles across the track.