Insights Map: AI-Powered Taxi Intelligence Platform
Improving Fleet & Driver Shift Operating Efficiency via a Next Generation Taxi Operating System
Street-hail taxis are a cornerstone of Hong Kong’s public transit network, serving over 60% of daily taxi journeys and providing essential point-to-point mobility. However, unlike app-based ride-hailing, street-hail demand has historically been difficult to track and predict at scale. Consequently, drivers rely on intuition when searching for passengers, resulting in high idle time, wasted fuel, and long passenger waits, with idle cruising averaging nearly 10 hours per day across 24-hour vehicle operations.
Funded by the Smart Traffic Fund, the GSCI research team partnered with the local taxi industry to develop Insights Map, an AI-powered next-generation taxi operating system that bridges the gap between street-hail demand and taxi supply to enable faster passenger matching and more efficient fleet operations.
Team Members
- Prof. Hong K. LO — Co-Investigator & Advisor
- Dr. Zheng LIANG – Project Manager & Lead Researcher
- Dr. Weigang LU – Researcher & Machine Learning Specialist
- Mr. Helbert PAAT – Researcher & Data Analyst
Core Technology & Architecture
Insights Map is a real-time predictive recommendation platform. Continuously learning from over 2,000 active taxis equipped with smart connected taximeters, Insights Map segments Hong Kong into 350+ micro-zones, forecasts pickup demand every 15 minutes, and cross-references it against live empty-taxi supply. Rather than acting solely as a demand heatmap, the engine evaluates net trip profitability and delivers clear repositioning guidance to drivers—optimizing fleet distribution and advancing sustainable urban mobility across Hong Kong.
Key Features / Highlights:
-
Real-Time Telemetry Sensing: Streams spatial location data every 5 seconds and live trip statuses from over 2,000 active taxis. Custom algorithms correct GPS signal loss in dense urban canyons and tunnels, creating an accurate second-by-second operational overview.
-
Granular Citywide Micro-Zoning: Divides Hong Kong's road network into 350+ micro-zones informed by road network topology, land-use patterns, points of interest (POIs), and driver behavior, balancing precise backend modeling with an intuitive interface for drivers.
-
Spatiotemporal Deep Learning Forecasting Engine: Operates on a continuous 15-minute cycle, integrating deep learning architectures to fuse historical trip records with real-time contextual variables, such as weather conditions, day-of-week patterns, and special events, to forecast short-term street-hail demand.
-
Dynamic Supply-Demand Matching: Weighs forecasted passenger pickups against the live supply of empty taxis in each zone. By balancing fare revenue against idle cruising costs, the system outputs a single, actionable recommendation for each driver.
Impact & Metrics Comparison
- 90%+ Demand Forecasting Accuracy: Trained and validated on a curated dataset of up to 6.45 million actual taxi trips (with weekly active fleet analyses covering over 277,000 trips), the model achieves over 90% prediction accuracy across Hong Kong's busiest zones in 15-minute intervals.
- 30% More Trips & 15%+ Earnings Increase (Simulation Results): In a ghost-agent simulation using historical working-day data, where virtual agents mirrored real drivers' working hours and starting locations but were assumed to strictly follow Insights Map recommendations, agents completed 31.2% more trips and achieved a 15.75% daily earnings increase by minimizing idle cruising time between fares.*
- Shorter Passenger Waits & Greener Mobility: By resolving spatial mismatches between street-hail passengers and available taxis, Insights Map reduces passenger wait times during peak hours and shift changes, alleviates urban road congestion from empty-taxi roaming, and lowers vehicular carbon emissions—directly advancing the Smart Mobility goals of the Smart City Blueprint for Hong Kong 2.0.
*Note: Simulation results derived from historical working-day data. Real-world performance will be empirically measured during upcoming field tests.
Milestones & Status
- Q4 2026 Field Testing: Finalizing the real-time empty-taxi supply overlay and driver profitability profiles on the web/mobile app. A field-based A/B pilot launches in Q4 2026 to validate real-world earnings and wait-time improvements across active fleets.
Media Coverage
