Article Preview
Top1. Introduction
Various carmakers and IT companies have recently developed various levels of automation for self-driving vehicles (On-Road Automated Driving [ORAD] committee, 2014). A self-driving vehicle requires the development of various technological levels (Bacha et al., 2008; Leonard et al., 2008), for example, a sensor level for detecting the environment and pedestrians (Dolla ́r, Wojek, Schiele & Perona, 2012), a communication level to obtain information on the traffic and other vehicles (Hiraishi, Ohwada & Mizoguchi, 1999), a motion level to make decisions based on the information obtained from the sensor and communication levels (Aguiar & Hespanha et al, 2007; Dolgov, Thrun, Montemerlo & Diebel, 2010), a route-planning level to generate a route to a destination (Nilsson, 1980), and finally, a cognitive level to clarify the relationship between humans and vehicles (Sega, Iwasaki, Hiraishi, & Mizoguchi, 2011; Harada, Iwasaki, Mori, Yoshizawa & Mizoguchi, 2014; Harada, Mori, Yoshizawa, & Iwasaki, 2015; Yoshizama, Nishiyama, Iwasaki & Mizoguchi, 2017). Thus, a wide range of technologies related to not only the level of movement but also the levels of safety and comfort are indispensable to realizing a self-driving vehicle. Among such technologies, in this paper, we focus on a route-planning technology, and discuss how to realize the avoidance of traffic congestion for an environment in which self-driving vehicles are widely used.
It is well known that traffic congestion on highways is caused by the unintentional slowdown of drivers on curves and uphill roads. In general, because self-driving vehicles can be controlled to maintain a constant speed, it has been stated that the spread of self-driving vehicles will lead to a reduction in traffic congestion. However, vehicles often have to stop at corners and signals on urban roads. Thus, unlike on a highway, the effectiveness of self-driving may not be as great as expected. Furthermore, if we consider a situation in which self-driving vehicles will be mass-produced and widely distributed, a large number of vehicles driving on similar routes generated through similar route-planning methods may cause new types of traffic congestion. Therefore, a policy to resolve the congestion problem is necessary if vehicles are to coexist under an exceeded road capacity.
Some approaches to the problems of traffic congestion have been suggested, including, an analysis and improvement of the road itself through traffic engineering (Amudapuram & Kalaga, 2012), and traffic management through signal control in the ITS (Intelligent Transport System) (Rahane & Saharkar, 2014). We approached the problem of traffic congestion using a route-finding method. When traffic congestion occurs along a route, our vehicles generate a new route to avoid the congestion. We executed some experiments on the relationship between the total efficiency and diffusion rate of self-driving vehicles using our own self-developed traffic simulator. We placed a large number of vehicles within a certain range, and made all the vehicles move toward the same destination simultaneously. In addition, we changed the rate of vehicles executing the avoidance of traffic congestion to verify the effectiveness of the proposed avoidance method. We can regard this as the diffusion rate of self-driving vehicles. As a result, we were able to clarify the phenomenon in which a total efficiency peak occurs, and in which the efficiency decreases as more vehicles generate an avoidance route. Therefore, it is not always best to generate an avoidance route, and the decision to drive along the current route without generating an avoidance route becomes important in certain cases.