Self-driving electric taxi fleets could make transportation more efficient and sustainable, but only if vehicles can continuously make smart decisions about where to drive and when to recharge. Researchers at Chalmers University of Technology in Sweden have now developed a new optimisation method that enables near-optimal large scale fleet control while accounting for uncertainties that arise in real-world transport systems.
The development of self-driving taxi fleets is progressing rapidly, and is predicted to make passenger transport safer, more environmentally friendly and more efficient. According to a report*, shared, self-driving electric passenger transport on a large scale – so-called electric autonomous-mobility-on-demand – can reduce an individual’s cost per trip by up to 60 percent compared to traditional taxis, while at the same time reducing emissions per trip significantly. But managing such a fleet is extremely complex.
"With self-driving taxi fleets, imbalances in urban vehicle networks can quickly arise. In the morning, for example, everyone wants to travel into the city center, and in the afternoon it's the other way around. Self-driving cars must then travel empty to pick up passengers for the morning and afternoon commute. In the case of electric vehicles, occasional stops for charging are needed. However, where and how long to charge is a difficult question”, says Sten Elling Tingstad Jacobsen, who holds a PhD in Control Engineering from the Department of Electrical Engineering at Chalmers University of Technology in Sweden.
Optimising vehicle availability and charging
The mismatch between available taxis and travel demand can negatively affect traffic conditions in a city. Hence, without efficient fleet control strategies, autonomous mobility services may increase waiting times and queue lengths, worsening urban traffic congestion.
In his doctoral thesis, Sten Elling Tingstad Jacobsen develops data-driven methods for improving the operation of autonomous electric taxi fleets. His research combines machine learning techniques for predicting travel demand, with advanced optimisation models for fleet planning and management. In addition, he investigates how deep reinforcement learning, a form of artificial intelligence that learns through interaction and experience, can be used to make real-time decisions in complex transport systems.
The first part of the research is about how empty vehicles should be placed in the city to be ready when the next traveler needs them. The second part of the research focuses on the additional challenges that arise when autonomous taxi fleets are electric and require regular charging. Using a machine learning method known as Bayesian neural networks (BNNs), travel demand and charger availability in different parts of the city are predicted based on historical taxi trip data from San Francisco and Chicago. Unlike many machine learning models, BNNs also provide estimates of how uncertain their predictions are. This information makes fleet management more robust when real-world travel patterns differ from forecasts, for example during unusual weather conditions. The machine learning is used only to provide forecasts, while the decisions about where vehicles go and when they charge are made by a transparent optimisation model with clearly defined goals and constraints, so the AI supports decisions, rather than making them on its own. The study also compares three types of electric vehicles with different battery sizes and energy efficiency, and how this affects fleet performance and service availability.
“Because demand, traffic conditions, and vehicle availability are constantly changing, the overall problem quickly becomes too complex to solve directly. Our approach addresses this challenge by breaking the problem into smaller, manageable parts that can be solved efficiently and then combined to estimate the best overall fleet deployment across the city. In this way, we obtain a solution that is very close to optimum while remaining computationally practical”, says Sten Elling Tingstad Jacobsen.
Coordination of very large fleets
Autonomous taxi services are already operating in several cities in the United States and China, but managing large fleets in complex urban environments remains a challenge.
"Self-driving vehicle fleets still face difficulties in uncertain and complex traffic conditions, and interactions with human drivers. Our robust method can help contribute to improving traffic coordination and supporting more reliable and trustworthy autonomous mobility," says Balázs Kulcsár, Professor at the Department of Electrical Engineering and supervisor of the study.
A key strength of the method is its ability to scale to large fleets while still making decisions quickly. According to Sten Elling Tingstad Jacobsen, the approach is designed to remain effective even under the uncertainties that characterise real-world urban transport systems.
"Many existing studies address similar fleet management and logistics problems. What sets our work apart is that we explicitly account for the uncertainties present in a large city and optimise fleet operations accordingly. We have also developed and tested the framework for different types of vehicles, making it both flexible and robust. This enables autonomous vehicle fleets to respond more effectively to changing conditions while maintaining efficient and reliable transport services.”
More about the research
A Predictive Chance Constraint Rebalancing Approach to Mobility-on-Demand Services has been published in Communications in Transport Research.
Combined Stochastic and Robust Optimization for Electric Autonomous Mobility-on-Demand with Nested Benders Decomposition has been published in Transportation Research Part C: Emerging Technologies.
The authors of the article are Sten Elling Tingstad Jacobsen and Balázs Kulcsár at Chalmers University of Technology, and Anders Lindman, Volvo Cars.
The research has been carried out together with Volvo Cars.
Read more about the development of self-driving taxi fleets:
*Report from Stanford University: The Future Of The Automated Mobility Industry In 2026: An Updated Strategic Management Perspective
Transportation Research
Computational simulation/modeling
Not applicable
Combined stochastic and robust optimization for electric autonomous mobility-on-demand with nested Benders decomposition
12-Aug-2026
No conflict of interest