In the development of driving strategies for autonomous vehicles, many different techniques have been studied. Using traditional control techniques alone will mostly lead to safe actions in “nominal cases.” However, they may lead to wrong actions in “critical scenarios,” where non-linearities tend to play a more prominent role.
Siemens Engineering Service investigates the combination of model-based and data-driven techniques to develop enhanced autonomous driving strategies and improve the vehicle’s control, accuracy, safety and comfort performance. This white paper outlines how to improve control strategies using a model-based approach combined with data-driven or machine learning methods.
In this white paper about the development of autonomous driving strategies you will learn how to:
- Use methods such as ‘learning adaptive controls’, improving driving strategies of existing Controls with data
- Improve controls based on Machine Learning using Reinforcement Learning and Imitation Learning
- Combine model-based controls with real driving data gathered with an expert driver
Challenge for automotive OEMs developing driving strategies
Maintaining safety is considered the most important factor and motivation of autonomous driving (AD) and advanced driver assistance system (ADAS) development. However, even advanced model-based control design and driving strategies may fail to achieve safety in critical scenarios. Comfort, or the occupant’s perception of the vehicle performance in ADAS scenarios, is another challenge for automotive OEMs. Customers will only accept the ADAS functions if they experience a comfortable feeling, and do not urge to take over vehicle control.
This paper aims to exploit the advantages of both model-based and data-driven driving strategies for AVs to cope with these challenges.