SimLucy — digital twin of the HM test vehicle in CARLA

Couples the CARLA simulator with the Autoware autonomous-driving stack over a ROS 2 bridge, so the Intelligent Vehicle Lab's test vehicle can be developed and tested in simulation instead of on the road. Contains the Autoware sensor kit and vehicle model plus the CARLA-side Python scripts.

M.Sc. Projekt Angewandte Forschung 2, Intelligent Vehicle Lab, Hochschule München. Boris Tokic, supervised by Prof. Dr. Fabian Flohr, 2024. Full report: docs/Ausarbeitung.pdf (German).

simlucy_demo.mp4

Overview

CARLA (Unreal Engine 4) provides the world, Autoware runs the autonomy. They don't talk directly, so a bridge sits between them: the existing Carla-Bridge (a TCP/IP client to the CARLA server) and the Autoware-Bridge, which remaps Autoware's control commands onto the Carla-Bridge topics. Actors and sensors are spawned from bridge_objects.json. Recording runs in synchronous mode so sensor timestamps stay evenly spaced.

Sensor & vehicle model — Carla_t2/

Autoware workspace overlay, adapted from the TUMFTM Carla-Autoware-Bridge. The t2 naming and base models come from there (originally a VW T2); they were re-worked here for the lab's Mercedes S-Class.

Sensor kit: front + rear lidar (32 ch, 75 m, 20 Hz, modelled on the Ouster OS0), front + rear camera (modelled on the Basler a2A2448-23gcPRO), IMU and GNSS.

• carla_t2_common_sensor_launch — processing nodes: lidar concatenation with self/mirror crop, and camera_node_container.launch.py, which runs YOLO on the front camera.
• carla_t2_sensor_kit_description — sensor frames, poses and calibration (xacro + yaml).
• carla_t2_sensor_kit_launch — one launch file per sensor, started by sensing.launch.xml.
• carla_t2_vehicle_description — dimensions, mirrors, dynamics and the vehicle mesh.

The vehicle mesh (MercedesCoupe.dae, ~98 MB) is not included — it is a third-party model and is linked instead (details). It is visual only; the simulation builds and runs without it.

Results

Route following, stopping for obstacles and traffic lights, weather changes and rosbag record/replay all work.

Known issues: YOLO leaks GPU memory across restarts; the GNSS node won't start (no macro yet, so the IMU macro is reused); the camera is capped at 800×600 because the full 2448×2048 resolution lags CARLA.

Build & run

Instructions are in Carla_t2/README.md: install Autoware and CARLA from upstream, build the carla_t2_* packages, then run CARLA, the TUM bridge and e2e_simulator.launch.xml. Install steps are not duplicated here.

Layout

Path	Contents
Carla_t2/	Autoware workspace overlay (sensor kit + vehicle model)
carla_scripts/	CARLA Python clients: spawning traffic, recording camera/lidar, sensor tests
docs/	report, working notes, images

Attribution & license

Built on the TUMFTM Carla-Autoware-Bridge (LGPL-3.0); the carla_t2_* packages derive from it and keep that license. Uses Autoware (Apache-2.0) and CARLA (MIT). Vehicle mesh from cgtrader, not redistributed. Full list in ATTRIBUTION.md. Released under LGPL-3.0 (LICENSE).