Real-time motion planning framework for connected and automated vehicles: from theory to scaled experiments
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Traffic congestion has been persistently growing in the US from 1982 to 2020 since road capacity has not grown at the same pace as the population in urban areas. In 2019, traffic congestion in urban areas in the US caused drivers to spend an extra 8.7 billion hours on the road, purchasing an extra 3.5 billion gallons of fuel for a congestion cost of $190 billion. Global climate change has also resulted in the pressing need for improved energy efficiency and reduced environmental impact on transportation. Additionally, traffic safety is another growing concern. In 2018, there were 5.2M traffic accidents in the US, resulting in more than 38K fatalities and 2.2M people injured. Equipped with capacities of computing capabilities and advanced communication and vehicle technologies, connected and automated vehicles (CAVs) provide novel and innovative opportunities to significantly reduce energy consumption, greenhouse gas emissions, and travel delays while improving passenger safety. ☐ This dissertation goes through a journey from designing an optimal decentralized coordination framework for CAVs at different traffic bottlenecks to implementing and validating them in a scaled testbed. The dissertation consists of three main parts. In the first part, we develop two different bi-level approaches based on a scheduling and recursive algorithm for coordinating CAVs at multiple adjacent signal-free intersections. In the upper-level planning, each CAV computes the arrival time at each intersection along its path while ensuring lateral and rear-end safety. Given the output of the upper-level planning, in the low-level planning, we formulate an optimal control problem for each CAV with the interior-point constraints, the solution of which yields the energy optimal control input (acceleration/deceleration). Considering a signal-free intersection, we present a learning-based decentralized coordination framework consisting of a hysteretic Q-learning combined with a FIFO queuing policy to minimize travel delay and improve fuel consumption while ensuring rear-end and lateral safety. ☐ In the second part, we provide three different approaches that can extend our framework to consider uncertainty and address it appropriately. We first present a single-level coordination framework for CAVs, in which each CAV computes the optimal unconstrained control trajectory without activating any of the state, control, and safety constraints.We integrate a replanning mechanism into our coordination framework, which can be implemented in a time-driven or event-driven fashion. This embedded replanning aims at introducing indirect feedback into the coordination framework to respond to the unexpected changes in the system to some extent.Using the theory of the job-shop scheduling, we further enhance our decentralized coordination framework by introducing a priority-aware resequencing mechanism, which designates the order of decision making.We formulate a robust coordination framework by including the deviations from the nominal trajectories as uncertainty. We then employ Gaussian process regression to learn the uncertainty from the possibly noisy observation of CAVs' time trajectories. After obtaining the statistical knowledge about the deviation from nominal trajectories, we construct the confidence interval for time, position, and speed trajectories.Finally, we enhance the framework by employing control barrier functions to provide an additional safety layer and ensure the satisfaction of all constraints in the system.Using the proposed coordination framework in the motion planning module, each CAV first uses simple longitudinal dynamics to derive the optimal control trajectory without activating any constraint. We require a vehicle-level controller to track the resulting optimal trajectory in a real physical system. However, the system's constraints may become active due to the inherent deviations between the actual trajectory and the planned trajectory. We address this issue by introducing a barrier-certificate module based on more realistic dynamics as a safety middle layer between the vehicle-level tracking controller and physical vehicle to provide a reactive mechanism to guarantee constraint satisfaction. ☐ In the last part, we introduce the Information and Decision Science Lab's Scaled Smart City (IDS3C), a robotic scaled (1:25) testbed capable of safely validating control approaches beyond simulation in applications related to emerging mobility systems such as coordination of CAVs. Then, we demonstrate the effectiveness of coordination of CAVs at a multi-lane roundabout and show its scalability in a corridor consisting of a roundabout, an intersection, and a merging roadway. ☐ These research contributions together result in a mathematically rigorous framework for the online coordination of CAVs in different traffic scenarios. This dissertation advances the state of the art in utilizing CAVs in real-world traffic scenarios to alleviate congestion, improve traffic throughput, and increase passenger safety.
Description
Keywords
Decentralized optimal control, Emerging mobility systems, Intelligent transportation system, Motion planning, Cooperative control