With a career spanning artificial intelligence, data analytics and advanced automotive engineering, Marius Reuther is helping to shape the next generation of ADAS, autonomous-driving and software-defined vehicle (SDV) technologies.
Before becoming CEO of b-plus automotive GmbH, Marius co-founded Incenda AI, pioneering innovative approaches to AI-driven data quality and processing for the automotive industry. Following its successful integration into the b-plus Group in 2023, he has led the expansion of intelligent, data-driven engineering solutions that enable safer, smarter and more efficient vehicle development.
Presentation
Closing the Real-to-Sim Gap – Turning Highway Drives into Simulation-Ready Datasets
Modern ADAS/AD programmes invest heavily in collecting and labelling real-world driving data, yet many projects stall when the dataset must be transferred into simulation for scenario variation, closed-loop validation and regression testing. The root cause is often not a lack of data, but a lack of dataset completeness: missing calibration and time-synchronisation evidence, inconsistent metadata, unclear labelling specifications and non-traceable quality KPIs. As a result, simulation teams cannot reliably convert real logs into a synthetic environment, even when the original data-collection and labelling budgets were substantial.
In this presentation, b-plus and IPG showcase an end-to-end real-to-sim-ready workflow using a Highway Drive Pilot as a reference case. The process begins with an ODD and labelling specification tailored to highway driving and perception tasks. Data is collected with a synchronised sensor stack that ensures time-exact recording, calibration integrity and structured metadata.
The dataset is then processed through high-volume automated labelling—including 2D and 3D bounding boxes, traffic signs and lane lines—reaching approximately 90% initial quality at very high throughput, with up to approximately 4,000 km processed overnight. b-plus then applies an independent quality-assurance and enrichment process established with TÜV SÜD since 2020, adds premium labelling tasks where required, and raises quality to approximately 99% with transparent KPIs and reports.
Finally, the curated dataset package—raw streams, labels, KPIs, reports and metadata—is transformed into a simulation scenario for CarMaker. The presentation demonstrates how a governed data pipeline removes friction at the real-to-sim interface and enables faster, more economical and higher-confidence validation loops for ADAS/AD development.
What the audience will learn
• Why many data-rich ADAS projects still fail at real-to-sim conversion, including metadata, calibration and KPI-traceability issues.
• How an ODD-driven pipeline produces CarMaker-ready datasets in days rather than weeks.
• How automated bulk labelling combined with independent QA enables scalable, cost-efficient and high-confidence validation.
Key advantages
• Real-to-sim readiness by design, with calibration, time synchronisation and metadata captured as first-class deliverables.
• ODD and labelling specifications agreed upfront, avoiding relabelling loops and definition drift.
• Independent QA using a four-eyes principle and a TÜV SÜD-established process in place since 2020.
• A flexible full-service approach that allows customers to rent or lease a complete sensor stack instead of purchasing hardware.
• A standards-led approach using structured metadata and interoperability to enable downstream simulation workflows.
