AI-Driven Fleet Scheduling for Electric Cabs — optimized in real time

The Opportunity

Allocating EV cabs across a continuous stream of pick-up and drop trips is fundamentally different from scheduling conventional vehicles. Battery charge is not just a constraint — it is the primary driver of every allocation decision. A cab cannot be assigned a trip it cannot physically complete, and charging downtime must be strategically scheduled to minimize fleet idle time. The operator managed a mixed fleet serving corporate campuses with strict SLA commitments — fixed pick-up windows, escort protocols, occupancy limits, and driver shift boundaries. Manual dispatch was reaching its limits: dispatchers juggled battery levels, driver hours, vehicle types, and campus rules simultaneously, leading to suboptimal utilization, excessive dead miles, and inequitable workload distribution across the fleet.

• 01Battery sufficiency for full journey.
• 02Driver working hours — every allocation within scheduled shift.
• 03Vehicle type compliance and campus-specific designations.
• 04Escort trips originate from campus; cab returns immediately after drop.
• 05Company-mandated occupancy limits respected per trip.
• 065-minute pick-up buffer for boarding.
• 07Equitable distance distribution across the fleet.
• 08Mid-schedule disruption minimization on cab unavailability.
• 09Intelligent fast/slow charging scheduling.
• 10SLA commitments to corporate clients requiring on-time performance within tight pick-up windows.

The Solution

We combined mathematical programming with advanced metaheuristics — engineered to converge on globally optimal allocations within the operational tempo of a live cab fleet. The three algorithms were chosen for their complementary strengths: Genetic Algorithms explore large, discontinuous solution spaces efficiently; Tabu Search refines promising regions through memory-based local search that avoids cycling; and Simulated Annealing provides the probabilistic escape mechanism needed to avoid premature convergence on local optima. The hybrid engine evaluates thousands of feasible schedules per optimization cycle, scoring each against a weighted objective function that balances utilization, dead miles, driver equity, and SLA compliance.

• 01Genetic Algorithms — evolutionary search across large solution spaces.
• 02Tabu Search — memory-based local search avoiding revisits to explored solutions.
• 03Simulated Annealing — probabilistic moves to escape local optima and converge globally.
• 04Weighted multi-objective function balancing utilization, dead miles, driver equity, and on-time SLA performance.
• 05Real-time re-optimization triggered on mid-schedule disruptions — cab breakdowns, charging delays, or trip cancellations handled without full schedule rebuild.
• 06Charging strategy optimization — intelligently sequencing fast and slow charging to maximize fleet availability during peak demand windows.

The Impact

The scheduling engine replaced manual dispatch with fully autonomous, constraint-aware allocation — handling the real-world complexity of live telematics, dynamic repositioning, and multi-constraint optimization at the speed operations demand. Dispatchers shifted from building schedules by hand to monitoring and exception-handling, focusing their expertise where human judgment adds value rather than on routine allocation decisions.

• 01Battery-aware allocations eliminate mid-trip range failures.
• 02Equitable distance distribution extends fleet life and driver fairness.
• 03Disruption-resilient re-planning on cab unavailability or charging events.
• 04Dead miles reduced through intelligent repositioning and trip chaining — maximizing revenue miles per charge cycle.
• 05Dispatcher role elevated from manual schedule-building to strategic oversight and exception management.
• 06On-time SLA performance improved through buffer-aware scheduling and real-time re-optimization.