[Paper Review] Fleet management for autonomous vehicles: Online PDP under special constraints

Sahar Bsaybes, Alain Quilliot+1 more|arXiv (Cornell University)|Mar 30, 2017

Transportation and Mobility Innovations12 references4 citations

TL;DR

This paper proposes a dynamic fleet management framework for autonomous vehicles in closed-site environments, modeling tram and elevator modes as online algorithms to minimize total tour length. It presents competitive analysis and practical evaluation, showing that real-world performance significantly outperforms worst-case theoretical bounds, enabling adaptive mode switching for optimal daily demand response.

ABSTRACT

The VIPAFLEET project consists in developing models and algorithms for man- aging a fleet of Individual Public Autonomous Vehicles (VIPA). Hereby, we consider a fleet of cars distributed at specified stations in an industrial area to supply internal transportation, where the cars can be used in different modes of circulation (tram mode, elevator mode, taxi mode). One goal is to develop and implement suitable algorithms for each mode in order to satisfy all the requests under an economic point of view by minimizing the total tour length. The innovative idea and challenge of the project is to develop and install a dynamic fleet management system that allows the operator to switch between the different modes within the different periods of the day according to the dynamic transportation demands of the users. We model the underlying online transportation system and propose a correspond- ing fleet management framework, to handle modes, demands and commands. We consider two modes of circulation, tram and elevator mode, propose for each mode appropriate on- line algorithms and evaluate their performance, both in terms of competitive analysis and practical behavior.

Motivation & Objective

• To develop a dynamic fleet management system that adapts transportation modes (tram, elevator, taxi) to changing daily demand patterns in closed sites.
• To model the VIPAFLEET system as an online metric task system with time-dependent subnetworks and vehicle repositioning.
• To design and analyze online algorithms for tram and elevator modes that minimize total tour length under real-time request arrivals.
• To evaluate both theoretical competitive ratios and practical performance of the algorithms on synthetic test instances.
• To provide a framework for clustering demands and subnetworks to enable globally efficient scheduling.

Proposed method

• Model the transportation network as a metric space (V, d) derived from a connected graph G, with stations as nodes and shortest-path distances as edge weights.
• Define customer requests as 4-tuples (t_j, x_j, y_j, z_j), specifying release time, origin, destination, and passenger count.
• Formulate tasks τ_j = (t_j, x_j, y_j, z_j, G') for online dispatching, where G' is an active subnetwork during time interval [t, t’].
• Propose the SIR algorithm for tram mode on circular networks, and SIF_M, SIF_E, SIF_L variants for mode-specific request handling.
• Design the MAIN algorithm for elevator mode on linear networks, supporting dynamic repositioning and request reassignment.
• Use competitive analysis to bound worst-case performance, and validate with computational experiments on varied request scenarios.

Experimental results

Research questions

• RQ1 How can online algorithms be designed for autonomous vehicle fleets operating in tram and elevator modes under dynamic, time-varying demand?
• RQ2 What competitive ratios can be theoretically established for the proposed online algorithms in minimizing total tour length?
• RQ3 How does the practical performance of the algorithms compare to their theoretical worst-case bounds in real-world-like scenarios?
• RQ4 What network and demand clustering strategies maximize efficiency across different time-of-day scenarios (morning, evening, lunch)?
• RQ5 How can a dynamic fleet management system intelligently switch between transportation modes to optimize economic and operational objectives?

Key findings

• The SIR algorithm for tram mode on circuits achieves a competitive ratio of Cap·|C|, where Cap is vehicle capacity and |C| is circuit length.
• The SIF_M and SIF_E algorithms achieve optimal competitive ratios of 1 in morning and evening scenarios, respectively, when requests are confined to specific subnetworks.
• The SIF_L algorithm for lunchtime scenarios achieves a competitive ratio of 2, reflecting the need for central coordination at the restaurant station.
• The MAIN algorithm for elevator mode on lines achieves a competitive ratio of 2·Cap·|L|, with practical performance significantly better than theoretical worst-case bounds.
• In computational experiments, the ratio TTL/OPT (actual tour length / optimal tour length) was consistently lower than the theoretical competitive ratio, indicating strong real-world performance.
• The framework enables effective clustering of demands into subproblems, allowing the use of tailored subnetworks and algorithms for improved global solution quality.