Project Overview

This document presents a design and implementation plan for an Autonomous Electric Vehicle (AEV) Ride-Hailing System. The primary objective is to develop a functional prototype that integrates core ride-hailing functionalities with the unique requirements of autonomous electric vehicles, covering the complete mobility-energy loop. Built on scalable, efficient, privacy-conscious cloud-native technologies, it leverages advanced algorithms for real-time fleet management. The system incorporates federated learning capabilities and seamless charging integration, delivering sustainable urban transportation. This comprehensive solution addresses next-generation mobility needs through energy-aware platform design and privacy-centric operations.

Key Differentiator: The system supports two dispatching approaches: a baseline Greedy algorithm and a more advanced, recommended Rolling-Horizon Min-Cost Flow (RH-MCF) algorithm for improved efficiency and robustness under load.
RH-MCF algorithm outperforms traditional greedy approaches by 10-20% in efficiency metrics while maintaining real-time performance through micro-batching and warm-start optimization.


Core System Components & Architecture

Architecture

The system is built using a microservices architecture for scalability and maintainability, orchestrated on Kubernetes. It comprises several key modules:

  • Customer Applications: Web portal and mobile app (React/React Native) for users to plan, book, and track rides.
  • Core Platform Services: A suite of backend microservices handling user management, trip lifecycle, AEV fleet orchestration (status, dispatch, routing, charging), reservations, payments, notifications, telemetry, and security.
  • Data Layer: Utilizes a hybrid approach with SQL databases (e.g., PostgreSQL) for structured data (users, trips, billing) and NoSQL (e.g., MongoDB) for high-volume telemetry and status updates. Streaming platforms (e.g., Kafka) and caches (e.g., Redis) manage real-time data flow.
  • Fleet Orchestration (ATMDM): The core AEV Management Module handles real-time vehicle status ingestion, information management, greedy or RH-MCF dispatch logic, maintenance/charging coordination, and route planning.
  • Energy Integration: Seamless integration with charging networks (e.g., via OCPP) for real-time bay availability, booking, and autonomous vehicle docking.
  • Privacy-Centric AI: A Federated Learning (FL) layer for training models (ETA, demand, energy, anomaly detection) directly on user/AEV devices, ensuring raw data never leaves the device.

Advanced Features & Technologies

The system introduces several groundbreaking capabilities and advanced features:

  • Rolling-Horizon Min-Cost Flow (RH-MCF) Dispatch: Recommended over the baseline greedy approach, RH-MCF batches requests and solves a global optimization problem to minimize total system cost (e.g., empty miles, wait times, energy use), leading to better fleet efficiency and rider experience, especially under high load.
  • Federated Learning (FL): Enables continuous improvement of key models (ETA, demand prediction) while preserving user privacy by training on-device and aggregating only model updates.
  • Autonomous Charging: Enables seamless automated charging capabilities.
  • Security & Safety: Comprehensive security measures including OIDC/JWT authentication, TLS encryption, RBAC/ABAC, field-level data encryption, and safety features like geofencing and remote E-stop.

Highlighting the benefits of these features:

Feature Description Benefit
RH-MCF Dispatch Min-cost flow optimization with micro-batching 10-20% lower empty miles, better SLA compliance
Federated Learning Privacy-preserving model training on-device Improved predictions without centralizing sensitive data
Autonomous Charging Seamless integration with charging networks Reduced operational downtime, energy optimization
Multi-factor Verification QR/BLE/on-device face matching for pickup Enhanced security while preserving privacy

Operational Workflows

The system implements several critical workflows to ensure smooth operation:

  • Trip Planning vs. Booking: A two-step process where users first "plan" to get quotes without locking vehicles, followed by an atomic "book" transaction that secures a vehicle and confirms the ride details.
  • Real-time Dispatch: Matches riders with the most suitable idle AEV using either the Greedy or RH-MCF algorithm, considering factors like proximity, battery range, and charging needs.
  • Pickup Verification: Ensures rider safety and prevents fraud using a multi-factor, privacy-preserving method involving QR codes, BLE proximity, and on-device face matching.
  • Autonomous Charging: After drop-off, the system directs the AEV to the nearest suitable charger, manages queueing, and initiates the docking procedure automatically.

Development & Deployment Plan

The project is planned for a 12-16 week development cycle, divided into five phases:

Phase 0
Foundations
(Weeks 1-2)
Phase 1
Core Trips & AEV
(Weeks 3-6)
Phase 2
Charging & Reservations
(Weeks 7-9)
Phase 3
Verification & Payments
(Weeks 10-12)
Phase 4
FL & Analytics
(Weeks 13-16)
  1. Foundations: Setup environments, CI/CD, and core infrastructure.
  2. Core Trips & AEV: Implement plan/book flows, basic ATMDM functionality, and live tracking.
  3. Charging & Reservations: Integrate charging network APIs and implement reservation logic.
  4. Pickup Verification & Payments: Complete rider verification workflow and payment processing.
  5. FL & Analytics: Deploy the initial FL orchestrator and analytics dashboards.

Each phase delivers working APIs, test suites, and operational documentation.

Expected Outcomes & KPIs

The system aims to deliver a superior ride-hailing experience with key performance indicators focused on:

  • Rider Experience: Fast match times, accurate ETAs, low cancellation rates, high NPS.
  • Fleet Efficiency: High vehicle utilization, low empty miles, optimized charging queue times, minimal SoC violations.
  • System Reliability: High API success rates, low latency, stable real-time communication.
  • Privacy & Trust: High FL participation, adherence to differential privacy budgets, secure pickup verification.

Highlighted KPIs

RH-MCF is expected to reduce empty miles by 10-20%
Average pickup ETA by 8-15% at moderate load
Significantly decrease SLA breaches (by 20-30%)
Charging queue penalties (by 10-25%) under high load

Competitive Advantages

  • Energy Integration: Full mobility-energy loop with autonomous charging
  • Privacy-Centric AI: Federated learning enables improvement without data centralization
  • Production-Ready Architecture: Microservices, containerization, and CI/CD pipeline
  • Scalable Dispatch: RH-MCF algorithm efficiently handles city-scale operations
  • Compliance Built-in: Security, privacy, and safety considerations integrated throughout

Strategic Importance

This system positions the organization at the forefront of autonomous mobility with a solution that balances operational efficiency, user experience, and sustainability. The platform's modular design allows for future expansion into multi-rider pooling, V2G integration, and city-scale simulation capabilities.

Recommendation: Implement the RH-MCF dispatch system as the highest-leverage improvement over traditional greedy approaches, delivering immediate efficiency gains while maintaining compatibility with existing infrastructure.

Future Extensions

Beyond the initial MVP, the system is designed to accommodate advanced features such as multi-rider pooling, auction-based dispatch, dynamic vehicle repositioning, energy-aware pricing, V2G participation, and integration with verifiable credentials for enhanced security and privacy.