Generation of New Driving Scenarios in CARLA Using OpenSCENARIO

Institute

Professur für autonome Fahrzeugsysteme (TUM-ED)

Type

Bachelor's Thesis / Semester Thesis /

Content

experimental /  

Description

Motivation

Scenario-based testing is central for validating autonomous driving and ADAS functions. CARLA supports standardized scenario definitions via OpenSCENARIO, which enables reproducible, shareable test cases. However, creating diverse scenarios manually is time-consuming and error-prone. This thesis builds a pipeline to generate new OpenSCENARIO files and execute them in CARLA at scale.

Goal

Develop a toolchain that can programmatically generate novel, valid OpenSCENARIO scenarios, run them in CARLA, and export a scenario set with metadata and basic quality checks.

Tasks / Work Packages

1. Survey & Setup

   • Study OpenSCENARIO (key entities: Storyboard, Entities, Actions, Triggers, Conditions)

   • Set up CARLA scenario execution workflow (runner / API)

2. Scenario Template Library

   • Implement reusable templates (e.g., cut-in, lead vehicle braking, pedestrian crossing, junction conflict)

   • Parameterize: spawn positions, speeds, routes, offsets, traffic density, trigger distances, weather/time

3. Scenario Generator

   • Create a generator that samples parameters and outputs new .xosc files

   • Ensure schema validity (XML validation / constraints)

4. Execution in CARLA

   • Automatically run generated scenarios in CARLA

   • Log execution status and key runtime information (collisions, min distance, TTC if feasible)

5. Dataset Packaging

   • Produce a scenario suite (e.g., 200–1000 scenarios) + metadata table

   • Provide replay instructions and documentation

Expected Deliverables

• A scenario generation toolkit (Python recommended) that outputs valid OpenSCENARIO files

• A scenario library (templates + generated scenarios)

• An execution script that runs scenarios in CARLA and stores results

• Final report + documentation (how to add new templates, how to validate, how to run)

Evaluation Criteria

• Validity: OpenSCENARIO conformance and successful CARLA execution rate

• Diversity: coverage across parameter ranges (speed, distance, road types, interactions)

• Reproducibility: deterministic reruns from random seed + config

• Scalability: runtime and robustness for batch generation/execution

Required Skills

• Python, Git, basic software engineering

• Familiarity with CARLA or willingness to learn

• XML / structured formats helpful (but not mandatory)

 

Work can begin immediately.
Please send me an email along with a current transcript of records and your resume, to: yuan_avs.gao@tum.de

Tags

AVS Gao

Possible start

sofort

Contact

Yuan Gao
yuan_avs.gaotum.de