Digital systems and software gain high importance in the automotive industry; we are facing a digital revolution. Ranging from assistance systems to full autonomous driving are key elements that will inevitably impact the success of car manufacturers around the globe. AI-based systems introduce a new dimension of services and capabilities to provide a new level of benefits to customers. The digital world demands for functions on demand, fast update cycles and high quality software tests, especially, for safety critical systems. Highly automated CI/CD pipelines operating in the backend are required that create together with the embedded world a hybrid system.
Our approach is based on MLOps to aim for automation and fast update cycles. AI models are trained in the cloud to leverage easy access to data lakes and vast system resources for training. For traceability reasons, all configured parameters are recorded. Finally, we use a tool chain to optimize, convert and compile the models that shall be inferenced in the vehicle. We offer different converters and compilers to optimize the models for different hardware / software stacks in the vehicle. This might be an ECU for body functions, an ADAS blade or an infotainment ECU. The models can be optimized for CPU-based execution as well as for accelerators such as GPUs or NPUs. The counterpart in the vehicle is an inference engine that loads, executes and monitors AI models on the dedicated ECUs as well as receives and forwards signals from and to other ECUs.
In this session, you will learn more about
- Automated cloud-based AI training pipelines
- Optimization and conversion technologies for enabling (embedded) in-vehicle inference
- In-Vehicle model inference and training
&nsbp;
