3 Questions for… Thomas Dannemann, Senior Director Product Marketing, Qualcomm
“GenAI can continuously improve ADAS algorithms”
Driving Innovation with On-device AI for Next-Generation SDVs: Dannemann’s topic on 22 May.
Qualcomm
In the run-up to the 6th International Conference Automotive Software Strategies, we asked Thomas Dannemann, Senior Director Product Marketing at Qualcomm, three questions. The event will take place on 21 and 22 May 2025 in Munich, Dannemann is one of the speakers.
As Senior Director of Product Marketing, Thomas Dannemann works with key automotive OEMs, Tier 1 suppliers, and other partners across the ecosystem to support the delivery of automotive solutions focused on infotainment and ADAS.
He has 20 years of experience working on technology solutions, primarily in the automotive industry. Dannemann holds a Master of Science in Electrical Engineering and Computer Science, as well as a diploma in Electrical Science.
ADT: What advantages does on-device AI offer over cloud-based approaches in SDV functionality?
Dannemann: On-device AI is essential for delivering real-time performance, privacy, and reliability in software-defined vehicles (SDVs). Running multimodal AI models, which combine vision, language, and sensor data, directly in the vehicle ensures split-second decision-making for driver assistance, personalised experiences, and intelligent services, even when cloud connectivity is limited. At the same time, developer-oriented software tools help streamline and simplify the process of deploying AI-driven applications. These provide an end-to-end workflow that integrates cloud-based tools with high-performance computing, enabling developers to create, test, and deploy advanced automotive applications more efficiently. While running AI in the cloud is possible, since the data sources come from the vehicle itself, keeping the data on the device ensures privacy, reliability, cost efficiency, and immediacy. By running local models with billions of parameters, vehicles can offer more personalised, responsive features.
How is Qualcomm Technologies supporting central-compute architectures in high-performance SDV applications?
Automotive architectures have evolved from purpose-built chips to centralised, cloud-connected processing that streamlines development and accelerates design cycles. These modern architectures rely on heterogeneous compute to handle sensor data fusion, domain integration, and real-time decision-making - from driver monitoring to environmental perception and automated driving functions. Multimodal AI algorithms process data from cameras, radars, and other sensors with minimal latency to react effectively to potential hazards and deliver seamless user experiences. This approach enables automotive developers to leverage transformers and end-to-end AI networks rather than adding them in isolated parts. This allows for scalability and growth, creating compound AI systems by combining various types of information (in-cabin and the surrounding environment) with the help of generative AI. As compound AI evolves, driving experiences will become safer, more personalised, and more efficient. For example: Generative AI can simulate countless driving scenarios and outcomes based on real-time environmental data and historical driver behaviour.
This capability allows advanced driver assistance systems (ADAS) to make context-informed adaptations about when to intervene and how to alert the driver, enhancing the driving experience. Using natural language processing and voice generation, generative AI can provide dynamic communication with the driver, offering guidance and alerts in a natural, conversational manner that adjusts to the driver’s behaviour and preferences. Generative AI coupled with large vision models (LVMs) can predict driver behaviour based on past patterns, enabling ADAS to anticipate potentially unsafe actions (like sudden lane changes or hard braking) and take preemptive actions to mitigate risks. As generative AI continues to learn from a wide array of driver interactions and scenarios, it can continuously improve the ADAS algorithms, making them smarter and more attuned to individual drivers over time. For long drives, generative AI can create personalised content such as audiobooks or music playlists. The content not only caters to the driver’s tastes, but also responds to their current level of alertness, helping to keep the driver engaged and alert. Qualcomm Technologies’ Snapdragon Digital Chassis brings together advanced connectivity, efficient AI-powered processing, and scalable compute to support these types of applications. It provides automakers with a proven, scalable technology suite for creating highly customisable, experiential, and upgradeable intelligent vehicles.
Where do you see the biggest hurdles in scaling AI-enhanced in-car systems?
Scaling AI-enhanced in-car systems requires a new approach to vehicle architectures. Delivering intelligent, personalised, and responsive driving experiences while meeting real-time performance and privacy requirements demands powerful edge-processing, scalable computing power, and a unified software environment. One of the biggest challenges is implementing architectures that can handle AI workloads across multiple domains, from driver assistance to immersive cockpit systems, all while maintaining low latency and high reliability. As compound AI evolves and grows, efficient scaling becomes critical. Real-time AI processing needs to happen directly on the device to meet performance, cost, and privacy demands, while cloud platforms are essential for large-scale data collection, training, and development. The Snapdragon Digital Chassis addresses these challenges by enabling the heterogeneous compute, AI acceleration, and connectivity needed for next-generation vehicles. Snapdragon Car-to-Cloud Services, for instance, help automakers manage and deliver new features, upgrades, and services over the air, enabling vehicles to improve and evolve over time while simplifying development and accelerating time to market.
