Hybrid Machine Learning

Vehicle dynamics control can benefit from incorporating data-based methods while not sacrificing knowledge and experience achieved with traditional methods.

Hybrid Machine Learning methods are based on data and physical models and offer a great chance to exploit the full potential in control, parameter/state estimation, and are helpful to close the Simulation to Reality gap.

Recent use cases considered by the research group:

• Health state estimation of vehicle chassis components
• Powertrain and lateral vehicle dynamics control based on deep reinforcement learning
• Modelling of motorcycle dynamics with different hybrid ML approaches
• Driver action prediction for ADAS systems
• Criteria and control of ride comfort applying data-based methods
• Modelling (semi)active dampers with hybrid ML approaches, e.g. universal differential equations (UDE)

Digital Twin Technology

The Internet of Things and cyber-physical systems are able to share experience and enable distributed learning among vehicles. Digital Twin Technology takes a key role in this context and is considered here to fundamentally extend model-based control. The connectivity between vehicles, infrastructure (e.g., cameras, traffic lights), and third-party providers (e.g., weather, maps) creates dense networks sharing real-time traffic, environmental and vehicle data. This wealth of data enables the creation of traffic Digital Twins (DTs) -- dynamic, virtual replicas of the physical traffic network.

While traffic DTs offer significant potential for simulating, monitoring, and optimising traffic scenarios, particularly for autonomous driving, contextual information, especially predictive insights from the vehicle environment (e.g. road friction, road trajectory), can benefit lower-level vehicle functions like chassis control, which is the main research focus of the research group.

The figures below show a specific use case on the application of cloud-based DTs in vehicle dynamics control to enhance lateral vehicle stability and vehicle safety.

^{Figure 1: Applied cyber-physical system used to enhance vehicle safety}

Figure 1: Applied cyber-physical system used to enhance vehicle safety

^{Figure 2: Real and digital twin with actuators used in the application case}

Figure 2: Real and digital twin with actuators used in the application case