SAFETY-CRITICAL PLANNING AND CONTROL FOR VEHICLES OPERATING IN DYNAMIC ENVIRONMENTS

Autonomous motion continues to drive engineering innovation. From autonomous vehicles to assembly line machines, maintaining safety while executing motion is paramount. Autonomous motion typically consists of two aspects, the plan and the execution. Path planners look at the given environment and find a safe, and oftentimes optimal, route for the system to execute the task. From there, a controller maneuvers the system along the planned path from the start to the goal state. However, if there are changes during the execution phase that cause the established path to be unsafe, the system must replan to account for the new environment. While these replanning procedures are increasingly quick, there is a period when the system either has to stop motion or risk executing an unsafe plan. In many applications, such as cars moving in the flow of traffic, the systems are not afforded the luxury of stopping to replan regardless of how quick the planner may be. This thesis introduces a framework that leverages safety-critical controllers to ensure safety in dynamic environments. In particular, we develop a Dual Cognition Motion framework which aims to establish safety buffers in the event of a changing environment to maintain safety while a global planner creates a new reference. To demonstrate this framework’s effectiveness, two demonstrations are presented using the 2-D unicycle vehicle model as an example system. Additionally, extensions of this work to 3-D environments, hyperparameter selection via a neural network, limited obstacle consideration, and risk-aware planning are presented. These extensions build on the existing framework and provide additional benefits to the motion planning and execution architecture. In total, we show that the Dual Cognition Motion framework provides safety guarantees in dynamic environments for systems with timing and motion constraints.