Many car drivers are surprised if they discover how much data their vehicles already collect about them, and increasingly determines actions such as braking, acceleration and also gear changing, if they have automatic transmission. This is only going to increase, for even if it is years or even decades before full autonomy becomes widespread, the technology will march forward at a faster and more predictable rate so that vehicles will be ready to roll as and when regulations allow.
Features of autonomy will infiltrate both driving and navigation in any case, testing the boundaries between privacy, safety, performance and insurance. Rather different issues and priorities arise according to the vehicle and journey type. Owners of fleets for example will be more intent on enforcing economy, while hire car firms will want to encourage driving that causes less wear on the vehicle.
All parties will be interested in safety but perhaps to varying extents. Then while fears over hacking of autonomous vehicles say to provoke accidents are almost certainly overblown and can be countered, there are more legitimate concerns over eavesdropping of personal data that may be stored on a car’s computer after synchronization with the mobile phones of drivers or passengers.
Then insurance companies want accurate feedback on driving behavior of individuals as well as associated with vehicles themselves in order to set competitive premiums within their risk management profiles. They want to identify genuinely safe drivers more quickly and less crudely than relying on time to establish discounts for no claims, which do not always accurately reflect the risk. Some vehicles are insured with several drivers, making it harder to identify the correct premium.
However, some insurers have reported success with “pay how you drive” policies that do not necessarily distinguish who is at the wheel but merely keep a running tally taking account of journey lengths and distances as well as behavior. If a policy is tied to a vehicle rather than individuals an insurer does not need to know or care who is driving, but just wants to measure the style at all times and calculate running premiums on the basis of aggregate behavior.
Driving style analysis has therefore become an important goal for on-board computers for various applications and has been the subject of numerous research papers, indicating that it is still very much work in progress. The first question is what parameters to measure and these can include average speed, fuel consumption, distance, number of braking actions and average acceleration over a given time.
Then the system should record the extremes such as heavy breaking, emergency stops and maximum acceleration when overtaking. The system can also record indicators from sensors that would be used in autonomous driving such as distances maintained from vehicles in front related to speed and how well a driver is keeping to lanes. The need for the vehicle to intervene frequently by nudging the steering wheel might indicate the driver was prone to drifting off at the wheel for example.
The objective then is to apply this data to assign individuals or, in the case of insurance, vehicles, to an appropriate profile. Or in the pursuit of safety the aim might be to encourage better behavior by issuing warnings when certain thresholds are exceeded, such as the minimum distance to the vehicle in front at a given speed. The advantage of statistical analysis based on Bayesian techniques, often referred to as machine learning, is that external factors as well as behavior can be taken into account when matching drivers or vehicles to profiles. Insurers for example might have different requirements than say fleet owners, wanting to combine past behavior or record of traffic offences or other relevant information with ongoing analysis to determine premiums.
The first step here is to define the profiles, which themselves will vary with application and can be adapted in response to data analysis. Then various analytical techniques can be employed to assign drivers or vehicles to the cluster that appears most likely to be correct and again this can be monitored and adjusted continuously in response to further input.
One more contentious aspect of the driver monitoring is concerned with enforcement of external rules such as speed limits or even control over routes. Currently many vehicles do provide warning lights or even sounds when speed limits are exceeded and, in some cases, slow down automatically. Ford is among manufacturers to have implemented such speed limiting technology in some standard models although it can be overridden temporarily by using the accelerator pedal.
Trials of various technologies under the banner of Intelligent Speed Adaptation (ISA) have been held, especially in Europe where the EU has conducted a variety of experiments as well as surveys of both consumers and stakeholders such as law enforcement agencies. The largest-scale trials have taken place in Sweden and the Netherlands with three types of ISA being compared. The least stringent is open ISA where drivers are merely given clear warnings when they break the speed limit. The second is half-open ISA where the warnings are reinforced by resistance on the accelerator pedal, while full closed ISA is where speed limits are enforced automatically with the driver unable to override it. However, the latter can either be mandatory, or voluntary when the driver can disable it.
Results so far show that all variants of ISA improve safety and reduce accidents with mandatory closed ISA not surprisingly having the greatest impact. However, while one quarter of EU drivers surveyed approved of some form of ISA there was a lack of enthusiasm for the full closed version which was deemed too intrusive at this stage. The preferred variant was half ISA.
That view also reflected some negative side effects of the speed limiting, which included frustration at having a limit imposed externally, compensation behavior by driving faster on road segments where the ISA system was not active and diminished attention by relying on the system to enforce limits and generally concentrating less.
These are probably teething troubles and would of course become irrelevant in the event of full autonomous driving. It may then be that, eventually, autonomous enforcement might allow speed limits to be raised as some advocates have suggested because that would still not impair safety if human drivers were not involved. It has been suggested that would make longer distance driving on highways more competitive with high speed rail.