Autonomous cars a long way off
TROUBLESHOOTER: Tech costs dropping, but self-driving vehicles still not ready for roads
Even though automakers have been relatively quiet at the early events of the 2018 auto show circuit (Detroit and Montreal), they seem to take advantage of every other media opportunity to boast about their progress with autonomous driving and aggressive timelines (think major automotive presence at Las Vegas’s Consumer Electronics Show).
Several makers are continuing to run pilot studies in warm, snow-free environments to demonstrate they will be more than ready for the road when these four-wheeled robots finally hit dealership lots. So, to provide some answers to questions consumers keep asking, we spoke with Grant Courville of BlackBerry’s subsidiary QNX.
Courville is a senior director at BlackBerry QNX’s Automotive and Embedded division and a self-professed car nut. QNX is Canada’s leader in autonomous-driving technology and has formed dozens of partnerships with hardware and automotive manufacturers, as well as software specialists working in this field.
Let’s review the different levels of autonomous driving. Level 1 represents limited driver assistance features. Examples are adaptive cruise control or lane-departure warning. Level 2 is partial autonomy where a computer controls steering, acceleration, and braking, leaving everything else to the human pilot. Common examples are automatic emergency braking and lane-keeping assist. Level 3 is conditional autonomy, where computers control many of the driving tasks, but will require human intervention from time to time (Tesla’s Autopilot for example). Level 4 is high autonomy, meaning a vehicle has the capability to complete a trip from point A to point B on most roads without any human touch (Ford and Volvo, among others, are promising this for 2021). Finally, level 5 is full autonomy on any type of road (from multi-lane freeways to gravelled cottage lanes (no carmaker has published hard time lines for this level to date and many analysts are looking toward 2030 before it hits the streets).
With these kinds of far-off dates, why should consumers be concerned about the technology?
First, we’re already using some of it on the roads today, so chances are good that the vehicles surrounding you on your commute have drivers relying on one or more assistance features. Examples include blind-spot detection, lane-depar- ture warning, adaptive cruise control, traction control and so on. Secondly, most if not all of these systems use software and components made by companies not owned or directly controlled by your carmaker, so knowing who makes what and what their reliability track records are, can be important to getting the best value. And finally, more than a few governments around the globe are concerned with the vulnerability of these systems to malicious software, so you should be as concerned about automotive viruses as you are about hacks on your other personal electronic tech.
We put three basic questions to BlackBerry QNX’s Courville: How well will autonomous systems work on Canadian roads? What kind of reliability can consumers expect? How will these vehicles interact with human-driven autos?
Almost all drivers of vehicles with lane-keeping assist features have experienced problems when road lane markings are worn away or snow covered, or their parking assist alarm continually goes off because of snow and ice covering the sensors. Current levels of driver assist can’t seem to handle these common problems and simply go offline when they are present.
Courville tells us that higher levels of autonomy rely on multiple systems to provide a safe level of redundancy. Instead of just visual cameras, these vehicle use a combination of radar, lidar (light detection and ranging), high-precision GPS units (accurate to within three centimetres), proximity detectors, hi-definition mapping software, and of course cameras.
In BlackBerry QNX’s own local trials in the high-tech sector of Ottawa, it has found good quality radar units to be particularly capable in navigating through rain, snow, and such. Sensor research and the complex software algorithms that they require, is one of the main reasons Gord Bell, co-founder of QNX, recently came out of retirement to rejoin the QNX team.
These autos will also benefit from electronic road network infrastructure, where they can receive information such as traffic-light timing, among other things. When it comes to more subtle things, such as determining a vehicle’s appropriate placement on a completely snow-covered rural road, on-board software and artificial intelligence come into play. This is where, as Courville puts it, systems can use visual cues such as ditch profiles, and hydro poles, (such as humans do) to determine the centre-line of the road.
Reliability has to be a major concern for drivers because the various components for these systems can be expensive to replace. (For example, a blind-spot radar detector on a Chrysler 300 lists for almost $1,000.) Courville is seeing prices drop for specific units as demand increases to critical mass points.
At one time, not long ago, radar units were considered cost-prohibitive for all but premium vehicles. Those prices are dropping. One of the current holders of the highcost component title, in Courville’s experience, are those high-precision GPS units which are used in multiple locations on prototype test platforms. BlackBerry QNX looks to use Canadian firms wherever possible, with better-than-average track records of long-life parts and a commitment to the automotive market. And they start, even at prototype and demonstration levels, by using automotive-grade hardware and BlackBerry QNX’s software to ensure their systems can stand up to real-world driving conditions.
How automated and ultimately autonomous autos interact with us plain human operators should be of concern to more than just early adopters because we will have to share a crowded road with them. For Courville and the BlackBerry QNX team, this is where their in-car software and artificial intelligence comes in. AI will permit high-level autonomous controllers to learn from every interaction on the road, and to adjust operation to predict less predictable organic driving behaviours. That technology is being refined with every kilometre of testing, including in the Ottawa valley and in the nastiest winter weather.