Real-world glitches mean we can expect a few more bumps on the road to autonomous vehicles
It has become a tenet of futurists’ faith that all new cars ultimately will be autonomous. That may be — probably will be — but it’s not going to happen in the very short timeframe envisioned by the most ardent boosters of those technologies, primarily people from outside the automotive industry.
From the very inception of the automobile, its technologies have evolved. There have been very few revolutionary changes and virtually none that became widespread immediately. Even such major advances as electric self-starters, automatic transmissions, air-conditioning and fuel injection took years, if not decades, to become mainstream.
In spite of much hype suggesting otherwise, such too is the probable fate of autonomous vehicle (AV) technologies.
Some have suggested — even promised — widespread availability of fully autonomous, Level 5 vehicles by 2020. But don’t count on it. Those really in the know suggest 2030 is a more realistic target, if even then.
Yes, there are some vehicles already on the market with claimed Level 4 capabilities in certain conditions, Cadillac’s SuperCruise, Mercedes-Benz’s Drive Pilot and Tesla’s AutoPilot among them. Plus, there are now thousands of prototype Level 5 systems undergoing testing in real-world conditions all around the globe, by traditional automakers and silicon valley interlopers alike.
But that last step, from Level 4 to Level 5, is a huge one. It necessitates expanding the capabilities from some autonomous functions in some conditions and some locales to fully autonomous functioning in all conditions, everywhere. A tall order indeed!
The experience gained with real-world test vehicles will be critical in determining just how quickly autonomous technologies are adopted. But that timetable came face-to-face with reality recently in the wake of two highly-publicized, fatal crashes that seem to have resulted from problems with their autonomous or semi-autonomous systems.
As for the practice of testing such systems on public roads, these crashes have cast a damper on that aspect as well, prompting Uber, as well as Toyota, to suspend real-world testing immediately.
The first of those involved a Volvo XC90, modified and operated by Uber, which struck and killed a pedestrian in Tempe, Arizona, while operating in autonomous mode. That vehicle was fitted with Uber’s own AV technology. Reportedly, Volvo’s built-in pedestrian identification and warning system, which is now well proven, had been disabled on the vehicle.
The second fatal crash involved a production Tesla Model X, which struck a concrete lane divider in Mountain View, California, while operating on Autopilot — a semi-autonomous system that requires some driver attention and intervention to control, although it hasn’t always been marketed in that way.
Initial analyses suggest that the Tesla’s system followed the wrong painted highway lane marking, directing the vehicle into rather than around the barrier. It had reportedly done so previously in the same car on several occasions and multiple similar incidents have since been reported by other Tesla owners, necessitating operator input to avoid a crash.
These crashes raise some serious concerns, both about the systems themselves and the wisdom in allowing them to be activated on public roads.
As for the systems involved, the two appear to have one commonality: an attempt by their makers to minimize the costs of the hardware required. Most players in the AV space, auto manufacturers and others alike, seem to agree that multiple Lidar sensors are essential components of the systems.
Lidar transponders send out thousands of laser light signals per second and measure how long it takes for them to bounce back, thus enabling the creation of high-resolution 3D images defining the car’s surroundings, even in the dark or in adverse weather conditions.
While most AV systems, including earlier iterations of Uber’s technology, incorporate several Lidar transponders, the system on the crashed Uber Volvo employed only one. The Tesla Model X has none, relying solely on a combination of radar and camera-based technologies to define its surroundings.
The limitations inherent in such systems raise obvious questions. How, for example, if such a system can’t clearly identify the correct lane from clearly painted lane markings, can it be expected to do so on roads covered with snow or ice?
As for the practice of testing such systems on public roads, these crashes have cast a damper on that aspect as well, prompting Uber, as well as Toyota, to suspend real-world testing immediately. They also initiated at least a delay in U.S. legislation that would have cleared the way for more widespread AV testing.
All of which is a problem in itself, for extensive testing is what is required to prove the capabilities of the systems. It’s a real Catch 22 that must be resolved if autonomous vehicles are ever to become reality. Evolution takes time!
