Model Predictive Control for Constrained Navigation of Autonomous Vehicles

Abstract

AbstractAs autonomous vehicles become increasingly prevalent in our daily lives, new control challenges arise to ensure their safety and the safety of their surroundings. This work addresses these challenges by developing a suitable regulator that strikes a balance between different objectives. The first one is ‘safety’, which involves satisfying constraints and consistently avoiding obstacles. The second objective is ‘exploitation’, which aims to optimize the utilization of existing knowledge about the environment, reducing the overly cautious behaviour of guaranteed collision-free approaches. The third objective is ‘exploration’, which pertains to the ability to discover potential unknown areas while avoiding getting stuck in blocked regions. The design of motion planning algorithms for such systems requires carefully managing the trade-off between these requirements. Among the various approaches to dynamic path planning, discrete optimization methods such as Model Predictive Control (MPC) have gained significant attention. MPC excels in handling state and input constraints to ensure safety while minimizing a cost function defined by the user, enabling both exploitation and exploration aspects. By developing a suitable regulator and leveraging MPC approaches, this work aims to address the complex control challenges faced by autonomous vehicles and other safety-critical applications, ensuring a balance between safety, exploitation, and exploration.