Research Release v2.0 · Simulation-Based

NYC Taxi Zone Intelligent Recommendation System

An interactive research platform demonstrating offline reinforcement learning for taxi zone dispatching. Explore forecasts, compare policies, and simulate AI-driven decisions using real NYC TLC data.

297+
Tests Passing
5
Policies
10
Zones
24h
Forecast

NYC Taxi Zone Map

Interactive map of Manhattan taxi zones with demand levels. Click a zone to view details and run simulations.

Select a Zone

Click a marker on the map to see zone details.

No zone selected yet.

Low Demand
Medium Demand
High Demand

AI Decision Simulation

Configure inputs and run the AI pipeline to simulate a taxi dispatching decision. All results are simulation-based.

Scenario Configuration

Policy Benchmark Comparison

Comparison of all evaluated policies across reward, revenue, and utilization metrics.

Hourly Demand Forecast

Select a zone on the map to see its 24-hour demand forecast compared with historical averages.

Click a zone on the map to load the forecast chart.

Benchmark Dashboard

Key performance metrics from the v2.0 research release. All values from calibrated simulation evaluation.

1.487
Ensemble MAE
Best
$1,822
DQN Avg Daily Reward
Best Policy
8.88 → 3.11
Calibration Fare RMSE
Validated
274/15
Tests Passed / Skipped
Passing

System Performance Overview

Note: All metrics are from calibrated simulation evaluation. Not real deployment data.

Research Documentation

Explore the full research documentation, reproduction guide, and model cards.