Reliable and safe control and navigation for autonomous vehicles in dynamic urban environments

In this thesis is presented an algorithmic architecture for systematic risk evaluation, mitigation and management intended for autonomous transportation vehicles. The methods presented span low level control, trajectory tracking and multi-vehicle coordination. A task separation between low level steering control and trajectory tracking has been implemented to spread the design effort across two functional blocks. A robust low level controller has been designed, and a comfortable and flexible Model Predictive Controller (MPC) has been implemented for trajectory tracking. This controller has been associated with a supervision mechanism that monitors its performance in real time to evaluate the probability to underperform. When such a risk is identified, the speed of the system is adapted. The multi-vehicle coordination block fulfils the planning task. It is a decentralized, probabilistic optimization algorithm that is naturally risk-adverse. It has been made compatible with mixed-traffic scenarios with human drivers on the road. Results show that risks are monitored and managed across the whole architecture. Furthermore, easy to understand risk metrics are outputted to make the algorithms decisions understandable by the users and engineers working on the system. The work in this thus proposes systematic risk management techniques transposable to all autonomous vehicles systems. It has been tested in simulations and on test vehicles.