Driving in mixed traffic involves numerous interactions (Image Credits: Keith Shaw https://www.vtpi.org/avip.pdf)
The transportation system is on the cusp of a new technological revolution—the Connected Automated Vehicles (CAVs). With fast-tracked innovations in the automation sector and the burgeoning number of related research studies, it is no longer a question of whether CAVs should be implemented on public roads; but rather a question of how they will affect the future of the transportation system and, in general, the societal norms surrounding them. This technology has the potential to change not just traffic flow dynamics, but also the status quo social behavior in transportation systems; which have been unchanged for almost a century.
CAVs have gone through significant changes since their first official introduction at the DARPA Grand Challenge. They are envisaged to provide new levels of safety, mobility, and efficiency by taking human errors out of the driving equation. The CAV technology, in general, is deemed to have the potential to reduce, if not eliminate, crashes (and hence a vast majority of traffic delays), traffic congestion, energy consumption, and air pollution, while improving transportation accessibility, mobility, and even land use. These promising impacts, indeed, justify the accelerated rise in the related research and development efforts in academia and industry to bring the idea of automated vehicles to fruition. CAVs are now able to accurately sense their local environment, detect and classify objects, interpret changes in the surrounding driving environment, and perform complex maneuvers accommodating associated safety regulations. Yet, numerous critical hurdles still remain, impeding the full (or even partial) operation of these vehicles on public roads beyond the testing phases.
The operation of automated vehicles hinges on standard algorithms developed for robotic applications. CAVs (also called robotic cars) are, in essence, automated decision-making systems designed to perform driving tasks in a connected automated driving environment. Vehicle navigation algorithms utilize a series of input sensory data, along with prior knowledge about the surrounding environment, and decide on the best-controlling action to govern the vehicle's motion towards its goal destination. Thanks to recent advances in computation capabilities, sensing, and navigation technologies, the problems of vehicle localization, mapping, control, and route selection in stationary environments around inanimate obstacles seem to be largely solved. Simultaneous Localization and Mapping (SLAM), particle filters, and the well-known A-star algorithm are all among the tractable solutions that have enabled CAV navigation in non-interactive environments. Open-source systems are now widely available to realize a level of automation where the vehicle can follow a given route, while accommodating some simple safety considerations, such as how to stop when an obstacle is detected by the range sensors.
Image Credits: ON Semiconductor https://www.fierceelectronics.com/components/autonomous-vehicles-mix-opportunity-and-uncertainty
However, the problem of navigating CAVs on urban roads goes beyond dealing with inanimate obstacles, sensors, and maps. It is predicted that despite the rapid development of CAV technology, human-driven vehicles with no/limited communication and automation capabilities will dominate the vehicle fleet for decades. According to the studies, traditional vehicles will still contribute to nearly 50% of the vehicles by 2030. Hence, CAVs, human-driven vehicles, pedestrians, and other users of the road will coexist for a significant period during an inevitable transition phase. That being said, CAVs should compete for time, space, and right of way with agents who are at least as intelligent and rational as they are. An interesting field experiment in Greece and France indicated that programming CAVs to stop for any obstacle in such environments can put them at a disadvantage: other road users soon learn to exploit CAVs’ caution and come to expect them to stop at any interfering scenario. Inefficient operations of CAVs are then inevitable, as the vehicle might get stuck waiting for others to take the right of way.
Understanding and predicting other agents’ behavior, especially when it includes predicting their belief about oneself, is a massively more complex problem than localization, planning, and vehicle control around non-interactive obstacles. Proximity to humans requires a form of "human-CAV negotiation" to ensure safe and efficient operations in human spaces. CAVs in such environments should effectively interact with nearby human actors, predict their actions, and behave in a socially predictable way. It is, therefore, of paramount importance to not only equip CAVs with the required techniques to safely and efficiently operate in mixed traffic environments.
Unfortunately, accurate predictions of human behavior remain a challenge in autonomous driving, and existing approaches often fail to provide reliable approximations. The key to an accurate prediction is to understand and characterize the interactions among humans and incorporate the underlying decision-making mechanism into the CAVs’ motion planning algorithms. This task is particularly challenging since any change in the behavior of a human actor can be in response to either other road users’ actions, or in direct response to the CAV’s maneuvers. For instance, while planning for a lane-changing maneuver, a CAV should predict the movements of the vehicle in the target lane (i.e., the vehicle that is directly impacted by the lane-changing maneuver in the destination lane) in response to its current leader, and in response to the CAV’s lane-changing trajectory. Understanding the decision-making mechanisms behind these interactions is critical to understand humans’ responses to CAVs’ trajectories.
Since direct human-vehicle communication is not often possible, CAVs should resort to algorithms that use available sensory data to predict the future form of the environment and plan accordingly. Also, further research is required to analyze the formation of collaborative behavior among humans and CAVs and how it might potentially change social norms over time. Answering these questions is a critical step towards enhancing the realism of simulation frameworks, and can facilitate the planning for future transportation systems.
Comments