Welcome to the METS-R SIM!

The METS-R SIM was initially developed for the project Multi-modal Energy-optimal Trip Scheduling in Real-time. Through extensive development, METS-R has evolved into a comprehensive road traffic simulator capable of scaling to large road networks. Similar to existing traffic simulators such as SUMO and VISSIM, METS-R employs car-following and lane-changing models to accurately simulate vehicle interactions, providing user-friendly online APIs for monitoring and manipulating vehicle and service behaviors.

Distinct from other simulators, METS-R Simulator uniquely:

  1. Introduces a structured framework for shared mobility services, including ride-hailing and microtransit, driven by intelligent agents representing geographic zones, service requests, and detailed travel plans for vehicles and passengers.

  2. Utilizes Kafka for real-time simulation of data streams generated by connected vehicles.

  3. Integrates seamlessly with Scenic for sophisticated scenario generation and testing.

Key Highlights

Shared Mobility Framework

METS-R provides native support for ride-hailing and microtransit (bus) services, with intelligent zone-level demand agents managing trip generation, vehicle-passenger matching, and ridesharing. Unlike SUMO or VISSIM, no third-party plugin or external control loop is required.

Real-Time Data Streams via Kafka

Connected-vehicle data streams (Basic Safety Messages, link energy/travel-time telemetry) are simulated using Apache Kafka, enabling realistic testing of cloud-based intelligent transportation algorithms under data latency and bandwidth constraints.

Fully Reproducible Simulations

METS-R supports saving and restoring complete simulation snapshots via the save() and load() interactive APIs. Any simulation state can be checkpointed to disk and replayed exactly, enabling rigorous scientific reproducibility and deterministic experiment comparisons across different control algorithms.

High-Performance Parallel Execution

The METS-R HPC module orchestrates multiple simulation instances in parallel Docker containers, dramatically reducing wall-clock time for large-scale experiments. See Performance Benchmark for a quantitative benchmark comparison against SUMO.