Autonomous Vehicles

TASA ID: 2246

Driverless vehicles, or as they are called autonomous vehicles, are a major topic of interest these days.  A self-driving vehicle offers many advantages to the public, especially to the elderly and people with physical impairments who otherwise would not be able to drive. Then there is the issue that autonomous vehicles do not get drunk or become drowsy, and are not distracted by cell phones or children in the back seat. Thus they should, in principle, reduce the incidence of accidents and injuries to occupants. Another advantage suggested by the concept’s supporters is that by controlling speed and vehicle spacing patterns on highways, better overall fuel economy can be achieved and CO2 emissions reduced. 

So, who is working on autonomous vehicles and when can you expect to be able to buy one?  The answer depends on how you define an autonomous vehicle. Google has been actively promoting and testing vehicles that can drive themselves for the last several years. But they can currently only legally test drive their vehicles on public roads in California, Florida, Michigan, and Nevada; subject to speed restrictions and having the vehicles appropriately labeled as test vehicles. Other states are considering similar legislation. From an international perspective, starting in January 2015, driverless vehicles will be allowed to drive in - yet to be designated - cities in the United Kingdom, once again for testing purposes only. Other European countries, as well as Japan and China, have enacted similar legislation.

The situation changes if you want a vehicle that can steer itself while you are in the driver’s seat in case of an emergency. Mercedes-Benz is now offering their 2014 S 500 Class vehicle with a self-drive option costing $2,800-$3,000, in addition to the cost of the vehicle itself. The version sold in the United States automatically adjusts the steering wheel so that the vehicle stays in the center of the lane. Similar systems are now offered by Infiniti, Lexus and Acura. The option for the S Class sold in Europe also self-drives at speeds under 6 mph, and can apply the brakes to avoid a collision. Mercedes-Benz leads the pack of automotive companies in terms of test experience. The company says its research on autonomous driving goes back to 1986, and that by 1994 they had developed technology allowing a vehicle to change lanes, pass another vehicle, and maintain a safe driving distance between vehicles. Mercedes-Benz and other automotive companies, most notably BMW and Volvo expect to incrementally offer features in the next two years. Others including GM, Ford, Daimler-Chrysler, and the other major European and Japanese companies expect to see vehicles with full autonomous driving capabilities on the road by 2020.

The Google story is quite different and began in 1986, when an autonomous vehicle named Stanley won the 2005 Grand Challenge sponsored by DARPA. The vehicle completed a treacherous 7.32 mile course in six hours and 54 minutes. Stanley was developed by the Stanford University Artificial Intelligence Laboratory, and key members of the project later joined Google, along with robotics engineers from another competitor in the race - Carnegie Mellon - to develop the current generation of Google autonomous vehicles. The latest rendition of the Google vehicle has no steering wheel, accelerator or brakes so that the occupant must totally rely on the control system. Google has collected more test data in terms of miles driven on public roadways than even Mercedes-Benz, and they claim no accidents have occurred to date. The unique feature of the Google approach is the use of a device called a Lidar (laser radar) that uses 48 rotating lasers mounted on the roof of the vehicle to gather very accurate information about what is on the road and what the local terrain looks like. The Lidar was first used on the Stanley vehicle. The images are very data intensive, and data is processed on-board the vehicle and by a remote computer which are connected by a high speed data link. Images are constantly compared to data collected from Google’s street map database.   The Lidar is produced by a California-based company, Velodyne, and costs a staggering $70,000. When asked if the Lidar can be replaced, Google’s answer is that they are working on it. Recently, a German company named Ibeo, states that it can bring down the cost to $250 per vehicle when produced in high volume. Google’s future seems to be aimed at collaborations with automotive companies such as, the ones it now has with Kia and Hyundai.

For a vehicle to be truly autonomous, requires that at a minimum, the vehicle diagnostics know where the vehicle is and where it is going, as well as information related to what lane it is in, and where the other vehicles are,  in close proximity,  and what speed they are travelling. The vehicle must also be able to identify the status of traffic lights and identify road signs such as stop, yield or merge, and identify if bicycles, pedestrians or animals such as, deer or farm animals are on the road. The vehicle then must use this data to safely navigate toward its destination.

Enhanced GPS positioning is the key means to determining the position of the vehicle on its designated route. GPS by itself is not sufficient given that contact with a satellite GPS signal  may not always  be available at all locations due to weather conditions or local road terrain. By using information provided by on-board   linear and angular accelerometers, vehicle position can be accurately updated.   Google also concurs that GPS can be inaccurate and uses Lidar data to correct GPS errors. Route information is supplied by a map service provider such as Google or Nokia, where the data is downloaded to the vehicle several times per minute. The downloaded data presumably includes the locations of intersections and traffic lights located along the route, so that the vehicle knows ahead of time that an intersection is ahead and can calculate when the vehicle will reach it.

Lane identification and positioning are two important tasks that must be addressed by a robotic controls system. Lane positioning is determined usually by analyzing in real time where the vehicle is relative to the lane markers painted on the roadway, typically using two stereoscopic cameras located in the front of the vehicle, and making small adjustments to the steering wheel angle. As previously mentioned, this feature is now part of advanced cruise control systems offered by several automotive suppliers.  

Collision avoidance is also a major design issue for an autonomous vehicle. Sensors mounted on the front and side of the vehicle are typically used to detect the presence of other vehicles on the roadway, particularly those in close proximity that are approaching on  a trajectory that may lead to a collision. Both short and long range radar are now in use. For example, the 2014 Mercedes Benz S Class has short and long range radars on the front and rear of the vehicle and a short range radar on the sides. These sensors not only measure distance, but also the speed of approach and direction.  To complement the radar sensors are 12 ultrasonic sensors located on the sides and rear of the vehicle. Their function is to measure the short range proximity of objects and detect objects in back of the vehicle when backing up or assist in self-parking operations. Current generation systems simply alert the driver of a possible collision, and in some situations apply the brakes. The automotive companies have also implemented algorithms to eliminate false positive conditions and eliminate unwanted and distracting warning signals.

The autonomous vehicle is designed to take corrective actions by changing speed or direction. Mercedes Benz implements no hands driving at low speeds such as extreme traffic congestion in their 2014 S Class. Another important feature of radar and ultrasound is to identify crash events that cannot be avoided. Crash detection times would then be predictive and not responsive. Crash severity could also be better identified so that seatbelt pre-tensioners could be deployed much earlier and airbag systems deployed in a far more effective way than today on-board systems can do. Braking could also be initiated to reduce crash severity. The benefits may be even more important for side impact crashes where current generation detection times are limited to around 10 to 15 milliseconds and the coverage of the side by inertial sensors is limited to the region between the A and B pillars, whereas ultrasound sensors would respond even in extreme frontal or rear locations.

The other key task is identifying road signs, traffic lights and objects on the road, including pedestrians. United States patent US2014/0016826 A1 titled “Traffic Signal Mapping and Detection” was issued to Google in early 2014. It discusses the use of frontal stereoscopic cameras in identifying the status and location of traffic lights by scanning the pixels in real time and looking for red, yellow or green “blobs” in the pixel map and not confusing them with the tail lights of other vehicles. This is a very complex task in terms of pattern recognition. Not covered in the patent is applying the same methodology to detect people or other objects including fallen tree branches or truck tire carcasses. Google and every automotive company offering an autonomous vehicle must solve this problem.

There is no doubt that autonomous vehicles will also cause accidents in the future, but knowing what we do now, what are the likely causes? The list below contains likely candidates.

Software Related Issues- According to an article in MIT Technology Review, there are currently around 100 million lines of code in each vehicle sold today, including entertainment systems. To put this number in perspective, there are eight million lines of source code supporting the new F-35 fighter jet plane. How many additional lines of code will be required for autonomous vehicles? There are no good estimates, but it may be in the range of 10 million. The Stanley from the DARPA sponsored race only had 100 thousand lines of code, but its mission was far less complicated. The most likely major problem is software bugs and the recent revelation that faulty software malfunctions caused accelerators in the 2005 Toyota Camry LR to stick.  The Toyota problems arose because the overall software architecture did not comply with industry guidelines. The other problem is checking embedded algorithms for formulation mistakes or incorrect algorithm parameters.  

Map Data Errors – It is difficult to understand what types of errors may exist in the road maps compiled by Google, Nokia and others, but it is highly likely that they do exist.  Google now asks its customers to report mistakes on a hotline. A recent incident in Iowa where a Google map survey vehicle entered a street going the wrong way and caused an accident illustrates the problem. The cause of the error may simply be that a township installed a traffic light at an intersection after Google made their survey. However, if the data is being used by an autonomous vehicle, the results could be catastrophic.

Component malfunction – Presumably, the sensors and controllers associated with autonomous vehicles have diagnostic codes indicating a malfunction, and that this data can be used to generate an error signal that can be transmitted to the occupant; as is the case with other systems such as, an airbag system failure. However, the problem could be that, for example, a camera or radar unit is out of alignment because of a minor collision. Accumulated dirt on the front of sensors can also impact performance. Here again, the autonomous vehicle is much more susceptible to error than a vehicle with a driver.

Unforeseen Events – It is virtually guaranteed that road events will occur which although possibly foreseen by the engineers, there may be no feasible solution to offer outside of having the occupant assume control of the vehicle.  A situation as simple as a policeman guiding vehicles along an unmarked temporary detour or a very recent bridge washout with no warning signs would most likely totally confuse the system. These types of events represent a major stumbling block to the current Google vehicle design. Other events may occur too quickly for the vehicle to respond such as, avoiding a vehicle running a red light at high speed at an intersection with limited lateral visibility.

What is not discussed is the impact that an intelligent roadway system might have on autonomous vehicles. If stop lights could communicate with the vehicle so that their status - at the time - when the vehicle reaches the intersection could be communicated to the vehicle, the control system diagnostics could be simplified.  Another facet of the intelligent highway is that vehicles could communicate with each other. These systems are technically possible, but given the financial problems of the Federal Highway Trust Fund it seems unlikely that it will be in place by the automotive company’s target date of 2020.

This article discusses issues of general interest and does not give any specific legal or business advice pertaining to any specific circumstances.  Before acting upon any of its information, you should obtain appropriate advice from a lawyer or other qualified professional.