Researchers at Sweden’s Chalmers University of Technology have developed an optimisation method designed to improve the coordination of autonomous electric taxi fleets in urban environments. Developed in collaboration with Volvo Cars, the approach combines machine learning-based forecasts with mathematical optimisation to manage passenger demand, vehicle movements and charging requirements while accounting for uncertainty in real-world conditions.
Autonomous electric taxis could make urban transport more efficient and reduce operating costs. A report from Stanford University indicates that autonomous taxi services have the potential to lower costs per kilometre by up to 60% compared with conventional taxis. However, coordinating large fleets remains a significant operational challenge, particularly when passenger demand fluctuates throughout the day.
Managing Demand and Charging in Urban Traffic
Rush-hour travel patterns create an imbalance in the movement of taxi fleets. In the morning, demand tends to concentrate on journeys towards city centres, while evening travel often shifts in the opposite direction. Autonomous taxis may consequently need to reposition themselves without passengers to meet demand in other areas.
For electric vehicles, these movements must also be coordinated with battery capacity and charging requirements. Poor planning can contribute to unnecessary vehicle movements, longer passenger waiting times and uneven demand across charging stations.
The Chalmers research team is addressing these challenges by combining machine learning with conventional optimisation techniques. To forecast passenger demand and the availability of charging stations, the researchers use Bayesian Neural Networks (BNNs) trained on historical taxi data from San Francisco and Chicago.
Bayesian Neural Networks Account for Uncertainty
Unlike forecasting methods that provide predictions without explicitly estimating their reliability, BNNs can indicate how uncertain their own predictions are. This capability allows fleet management systems to account for situations in which actual conditions differ from historical patterns.
For example, sudden rainfall may alter passenger demand and change where vehicles are needed. By incorporating uncertainty into its forecasts, the system is designed to remain robust when travel patterns change unexpectedly.
The AI component does not independently decide where vehicles should travel or when they should charge. Instead, its forecasts provide information for a mathematical optimisation model that makes planning decisions according to predefined rules and constraints.
Breaking Down Fleet Planning Into Smaller Tasks
Managing a large autonomous taxi fleet across an entire city involves numerous variables, including passenger demand, vehicle locations, battery levels and charging station availability. To handle this complexity, the algorithm divides the overall planning problem into smaller sections and combines their results into a coordinated fleet plan.
This approach enables the system to generate decisions in fractions of a second, even when dealing with very large fleets. Faster planning could help operators respond to changing demand while balancing vehicle availability and charging needs.
Vehicle Battery Capacity Also Affects Fleet Efficiency
The researchers also evaluated three types of electric vehicles with different battery capacities and energy-efficiency characteristics. Their findings indicate that vehicle technology can have a substantial influence on fleet availability.
Larger batteries may provide greater operating flexibility between charging sessions, while vehicle efficiency and charging schedules also affect how long taxis can remain in service. The results highlight the importance of considering vehicle specifications alongside demand forecasting and fleet coordination.
By combining uncertainty-aware predictions with rule-based optimisation, the research offers a framework for making autonomous electric taxi operations more reliable. The approach could help operators coordinate passenger services and charging infrastructure more effectively as autonomous mobility systems develop.