Blog: Should an autonomous vehicle (AV) be designed to detect and try to avoid rear-end collisions (i.e.
To understand this part fully you need to look at the first part of this series: https://medium.com/@ratularn2/should-an-autonomous-vehicle-av-be-designed-to-detect-and-try-to-avoid-rear-end-collisions-i-e-400adb68122c
Regarding the ethics setting taxonomy described in ETHICS SETTINGS FOR AUTONOMOUS VEHICLES by Jason Millar (2017), the rear-ended collision setting belongs to high-stack non-expertise-based ethics settings. Thus, it is a hard case. AV system engineers cannot solve the issue alone. Although they can improve their system and decrease the rear-ended chances, they need input from users and regulators to obtain a robust decision-making system while facing an inevitable collision dilemma. The following shows three different approaches to resolve or ease the rear-ended tension.
- Dissolve the Tension: In most rear-ended cases, the cause of the collision is due to the uncommon driving style of the AV. The self-driving system on the AV is still unmatured nowadays, and the designers focus more on recognizing and avoiding crashing on the cars and pedestrians in the front. To improve the self-driving system to be more ‘human-like’, the designers can use machine learning technology in training the driving style of the AV as well, just as they use it in object recognizing. They can also design the system to be able to learn the driving style of the car following it so the system will have more information input to avoid a rear-end collision. All of these require a huge amount of user data and behavior studies as an input. Therefore, training an AV to drive like a human driver is not an easy job, it may take years or even decades for our engineers to achieve this goal with the help of users and regulators. Nevertheless, once the AV can drive like a human being, the tension will be dissolved.
- Compromise: the designers may apply a compromise solution, implementing a more straightforward feature to avoid some of the rear-ended collision. The first idea is trying to find an escape path to avoid the collision. The second one is trying to maintain a proper distance and speed towards the vehicle behind. Either design has some limits, but it is a compromise made by the designers. Designers should also add a feature to display a warning sign at the back of the AV to explicitly inform the car behind when it is in autopilot mode. When drivers behind see this sign, they can either increase the distance between the AV and their vehicle or they can change the lane if possible when they feel uncomfortable. The regulators should also support this solution by making related regulations to eliminate the potential conflicts between the drivers involved in a rear-end collision. The driver of the following car has to understand the limitations of this solution and be more cautious when they follow an AV. Hence, everyone has to compromise in this situation.
- Tradeoff: Another way of solving this tension is by prioritizing some stakeholders and values over others by trading-off the values. In this case, I can prioritize the safety of the pedestrians who are in front of and near the AV over the safety of the driver behind. The reason is that pedestrians are more vulnerable in an accident. Therefore, the system designers will not develop the rear-end avoidance feature because it may cause potential harm to the pedestrians.
Prototype-based on the compromise resolution:
In the current regulation, the designers do not have to design a rear-end avoidance system because the following vehicle will take responsibility for the rear-end collision. In addition, the company does not implement this feature because they want to be cost effective. In my perspective, technologically we are not affluent enough to dissolve the tension totally. As I mentioned in our previous part that I think it will take nearly 1 or 2 decades to resolve all issues. On the other hand, we need at least some basic attributes in the autonomous car which can avoid rearward collisions in some simple situation. We do not want to deal with complex situations right now. Therefore, after critically examine every issue I think it will be perfect to work with the compromise resolution in this present condition.
Right now, the designers may make a compromise to implement some new feature to avoid some of the rear-ended collision. I find two ideas which can be suitable for the present days. One idea is trying to find an escape path to avoid the collision. The other one is trying to maintain a proper distance and speed towards the vehicle behind but both solutions have some limits and may introduce new risks to cars and people near the AV. On the other hand, from the perspective of the driver in the vehicle behind, they need to understand the limitations of these solutions and be more cautious when they follow an AV. Thus, they also make a compromise.
Unfortunately, these compromise resolutions cannot completely dissolve the tension. If a rear-end collision still happens, the regulators should intervene and ease the tension.
In this prototype, I describe a list of products and policies for three stakeholders mostly involved in the tension, AV system designer, the driver in the vehicle following the AV and the law enforcement department.
- AV system designer:
In the present time, due to getting a cost-effective model we should not totally focus on 360-degree full view system for autonomous vehicles. To get total 360-degree we need powerful sensors, LIDAR, Radar to get the data. In addition, to process these data we need fast processing unit, clear object recognition system. However, in the system designer perspective, here we proposed two solutions to get a more precise outcome. Furthermore, these two solutions also have some limits and may introduce new risks to cars and people near the autonomous vehicle. Therefore, we have to compromise something if we want to implement these two solutions.
Maintain a proper distance and speed towards the vehicle behind: Now, all the autonomous car maker only considers the spacing to the vehicles ahead of the autonomous vehicle, though they think to keep a safe distance with the following car is redundant. To avoid rear-end collisions, here I proposed almost the same system as fort-end because, in my perspective, it is a cost-effective and safe model. To implement this model, we need extra sensors for the backend, velocity-controlled system for rear-end, combined constant inter-vehicle spacing system (wireless communication) for both front-end and backward, LIDAR data from the rear-end, and extra processing power for rearward data processing.
In the figure above, I show a basic control processing unit for space and speed control. If we are able to implement this system, then we can at least measure a safe distance from the backward car, and we can control our constant speed so that the rearward driver does not feel uncomfortable.
Though, it is not always possible to ensure the comfort issue for the backward driver of the autonomous car after implementing this feature because sometimes the situation depends on the front-end car, traffic congestion, weather condition, and so on. First, if the front-end car does not obey basic code & conduct of driving and move awkwardly then it will hamper the overall performance of the and makes it start to move awkwardly as well. Secondly, this feature is difficult to utilize during the traffic congestion of some megacities like Toronto, Beijing, New York, Delhi, Dhaka, etc. Thirdly, during the bad weather condition, I am not sure how this system will perform. Finally, I think this feature mostly emphasize on the safety issue of the autonomous car for both the front-end and back-end but not fully guarantee the comfortable issue for the rear-ending driver.
Trying to find an escape path to avoid the collision: If the autonomous vehicle can sense any danger or awkward behavior from the backend vehicle through the backward sensors then it can immediately change its lane to ensure the safety of the passengers. To implement this feature in the autonomous vehicle we need extra sensors for the backend, two sensors for the left and right, LIDAR data from the rear-end, and extra processing power to run smart electric steering actuator. We need smart electric steering actuator to immediate to move right or left. The car will take the left lane if it gets a safe signal from the left sensor otherwise it moves right. If it gets a safe signal from both the sensor then it can move either way based on probabilities.
However, this system only works if the right or left lane is clear to move for the AV. In the following figures, we will show some other possibilities where this system will not work and raise some ethical issue.
We can easily change the lane for the situation ‘a’ if we sense any danger from the rearward but what will happen for situation ‘b’ and ‘d’. There is no ideal escape root for these two situations, and we have no answer how autonomous vehicles will behave in these situations. In my perspective, if it takes any wrong decision to hit another vehicle based on probabilities to minimize the risk then too many ethical issues will raise. In the situation ‘c’, it can take the left lane, but it is risky and uncomfortable for the pink car which comes from behind. To ensure the safety of its own it elevates the risk factor of that pink car and I think it is another ethical dilemma.
2. Driver in the vehicle following the AV
Most of the time, the rear-end happens because the driver in the following vehicle does not know the front car is an AV. So, he/she has no expectation of the uncommon movement caused by the autopilot system. In this prototype, I introduce design to inform the driver this autopilot information and it can let the driver be prepared. Each AV should install a display screen or a special light on the top or back of the vehicle. When the autopilot is enabled, the screen displays “AUTO MODE” or tuns on the special top light to warn the following driver. With this information, the driver behind will have a clue and get more cautious when he/she follows the AV. They can either increase the distance between the AV and their vehicle or they can change the lane if possible when they feel uncomfortable. Because the avoidance system on an AV those I proposed are not fully perfect. Therefore, the human drivers following must make some compromise to be careful themselves.
3. The Law Enforcement Department
Despite the compromise the designer and the following driver made, rear-end collision can still happen. At that situation, we need another stakeholder to intervene and ease the tension. The lawmakers should write some new regulations of the vehicle collision involved AV. The following is a list of possible policy:
a. If the AV is in autopilot mode, but it does not display this information to let other drivers know. The AV will take a part of the responsibility in a collision.
b. After a rear-end collision involved an autopilot AV, the police have the right to check the system data record stored in the AV because it is the “real driver” of the vehicle. The police should analysis both the traffic camera and the data to investigate the cause of the collision.
c. Make some regulations to prevent another vehicle follows an AV too close. They can define a distance required for other drivers to keep from an autopilot AV.
To balance both parts, the regulators should also limit the AV driving in autopilot in some traffic condition. For example, in downtown areas, school areas, during rush hours, during the night, AV cannot be in autopilot. Another possible rule is to limit the lane an AV can use. For instance, autopilot AV cannot use the fastest lane on the highway.
To sum up, I want to say that the proposed compromise resolutions are not perfect enough to resolve all the ethical issues right now. However, I try to provide some possible solutions which can be cost-effective and easy to implement. For the future, I have another model for the system designers and that model will be more efficient for the autonomous vehicle. In my opinion, that model can be able to dissolve most of the ethical issues for the rear-end, though it needs more processing power and it will increase the cost. Furthermore, our goal is to provide the most perfect model for the future, that’s why I think it is better to propose some model which will help the system designers to build some futuristic autonomous vehicle model which can mimic and adopt the human behavior aptly. Here, I only concentrate on rear-end collisions for the future model.
Youtube Link of the Project: https://www.youtube.com/watch?v=llxZhZdQ5SQ&t=2s
1. Millar, J. (2017). Ethics settings for autonomous vehicles. doi: 10.1093/oso/9780190652951.003.0002
2. Jitendra Shah, Matt Best, Ahmed Benmimoun, and Mohsen Lakehal Ayat. (2015). Autonomous rear-end collision avoidance using an electric power steering system. doi: 10.1177/0954407014567517
3. P. Shen, H. Zou, X. Zhang, Y. Li and Y. Fang. (2017). Platoon of autonomous vehicles with rear-end collision avoidance through time-optimal path-constrained trajectory planning, 2017 11th International Workshop on Robot Motion and Control (RoMoCo), Wasowo, 2017, pp. 232–237. doi: 10.1109/RoMoCo.2017.8003918