Customizing Your Self-Driving Car

In the future, intelligent transportation systems (ITSs) will involve connected vehicles, including driver-assisted vehicles and self-driving cars, as well as on-board mobile devices, sensors, and the software and algorithms that govern the functioning of these devices and their communications. Despite recent improvements, each year tens of thousands of lives are lost and billions of dollars are wasted because of traffic inefficiencies in the United States alone. Improvements in the transportation systems could have an enormous impact on lowering these statistics.

In this research, we aim to establish a new approach in design of safety systems, which is based on the individualization and customization of these systems to specific drivers and their environments. This means that wireless communication protocols, as well as algorithms that communicate to users, can be designed in an intelligent way in order to take advantage of all the statistical data that is available regarding the driver and his/her environment.

To accomplish this objective, we can use the technology to collect driver performance data and subsequently learn driver characteristics and driving strategies. This information, along with data collected from other vehicles and roadside units, can be used to customize the technology to each driver. With this, it is possible to adapt warnings or automatic control strategies to each driver. Meanwhile, vehicle-to-vehicle (V2V) communication can be dynamically tuned to make efficient use of finite bandwidth and guarantee the transmission of information critical to safety.

In this way, we should consider that there is an uncertainty of the message delivery between two specific vehicles, while other vehicles might also transmit simultaneously. Our research shows that by proper adaptation of wireless communication and warning algorithms, we can potentially reduce accident fatalities by a considerable amount.

To understand the benefit of V2V communication, consider a traffic stream where a chain of vehicles moves with same speed. When the first vehicle in the chain brakes, the driver of the following vehicle applies the brake after her perception reaction time (PRT). If no intervehicle communications are employed, vehicle Vi applies the brake after the sum of PRTs up to the driver i. With the communications, this time will change to the communications delay plus PRTs of the driver i. This is shown in Figure 1.

Hossein_Fig1new
Figure 1. Communications delay versus sum of PRTs, illustrating the time before a driver in a chain applies the brake

Some drivers may think that some of the received warning messages are not needed, because the drivers are aware of their own response time empirically and they know that they can react to stimuli fast. These warning messages are false alarms for these drivers. These warning messages may frustrate the drivers with an overly high number of false alarms, causing them to ignore warnings or even disable the system. To address this issue, we propose estimating the PRT of drivers and personalizing warning messages based on individual PRTs. Figure 2 shows that at the same accident probability for each driver, the false alarm rate can be reduced by at least 30% by employing the estimated individual distribution instead of the population distribution. Thus, it is of vital importance to minimize false alarms so that the system sends warnings only when they are needed.

Hossein_Fig2new
Figure 2. False alarm rate versus the probability of accident based on using average response time or individual

Now, we should determine how channel access probabilities of vehicles and vehicular communications can be adapted to drivers’ characteristics. In a network of vehicles, each vehicle transmits with a specific probability in the transmission medium. Large channel access probabilities lead the system to excessive interferences and, consequently, low probability of packets being successfully received (success probability), while very small values reduce the success probabilities since the probability of the favorite transmission is low itself. Therefore, there is an optimal value, given both the physical data obtained by vehicular networks and the communications protocol requirements, which results in lower collision probability of vehicles. We can find the expression of packet success probability in a network of vehicles based on channel access probability of vehicle.

We then use a recursive algorithm to tune the transmission probability of each vehicle based on the individual characteristics of drivers. The PRT of the driver, traffic conditions, and communications delay are three factors that play roles in assigning channel access probabilities to vehicles. In simple terms, we categorize the drivers into safe and unsafe drivers based on perception-reaction time. The unsafe vehicles are the ones whose drivers have long perception-reaction time and low distance to the vehicle in front. In other words, unsafe vehicles have higher collision probability. Then we assign different channel access probabilities to unsafe and safe vehicles respectively.

Figure 3(a) illustrates the collision probabilities when channel access probabilities are assumed to be equal for all vehicles. Figure 3(b) shows the scenario in which different channel access probabilities are assigned to unsafe and safe vehicles. The minimum collision probability in the second scenario improves by 25%.

Hossein_Fig3a
Figure 3(a). Collision probabilities when channel access probabilities are equal for all vehicles
Hossein_Fig3b
Figure 3(b). Different channel access probabilities are assigned to unsafe and safe vehicles.

Our simulation results confirm that unsafe vehicles need to inform other vehicles of their perilous situation more frequently than do safer vehicles. In other words, with higher channel access probability for unsafe vehicles, we can achieve lower collision probabilities.

Written by Hossein Pishro-Nik, UMTC Research Affiliate and Associate Professor in the Department of Electrical and Computer Engineering (ECE) at UMass-Amherst. This research was supported by the National Science Foundation under Grant CCF– 0844725 (PI: Hossein Pishro-Nik). It is a joint work with ECE PhD students Mohammad Nekoui, Ali Rakhshan, and Mohammad Kohsravi, and Professor Daiheng Ni from the UMass-Amherst Department of Civil and Environmental Engineering. For more information and access to published papers, please visit http://www.ecs.umass.edu/ece/pishro/publications.html.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s