The only way forward

The Alphabet subsidiary, Waymo, completed its first external round of funding on Wednesday, March 3, raising $2.25 billion. The company will use these investments to scale up research into its systems for driverless, autonomous cars.  

Waymo claims that vehicles it has integrated have done more than 32 million kilometres on public roads. It operates a robot-taxi hailing service in Mountain View, California, for its employees, as well as a fully public service in Arizona. Arizona has shut down Uber’s offering of driverless cabs following a fatal accident but Waymo continues to operate.

Waymo also has an agreement with Jaguar Land Rover to convert 20,000 electric I-Pace Jaguars into self-drives and it demonstrated its fifth generation self-driving equipment, the Waymo Driver, on these vehicles.

Creating an autonomous vehicle or converting one, are expensive tasks in terms of R&D. Autonomous vehicles have needs unique to them at the moment. They need massive quantities of data about their surroundings to act as their “eyes and ears”. They also need hardware and AI-driven software to process that data in real-time, and issue driving instructions.

A car has to read stop signs, directional signage, identify lane changing zones, and speed limit notices from a reasonable distance. It must do this even in poor weather conditions, and at night. A car must also detect other moving objects at near and far range, identify these, judge relative speeds, and adjust its own speed and direction accordingly.

Companies in this area tend to end up designing their own equipment, and fabricating most of it because off-the-shelf components don’t meet their requirements. LIDAR — Light Detection and Ranging Radar is one of those key components. LIDAR provides both long-range vision as well as short-range detection, which helps with avoiding other moving vehicles. Along with cameras and radar, LIDAR provides the coverage that autonomous cars require to navigate public roads.

Waymo’s new system, the Waymo Driver, has a LIDAR unit plus cameras mounted on the roof to provide 360 degree long-distance coverage. It also has four perimeter LIDAR plus radar units mounted at the four corners to provide short-range vision, and peripheral vision. It has a mindboggling total of 29 cameras as well.  

The long-range unit on the roof can identify pedestrians and stop signs from more than 500 meters away. The short-range units provide enhanced spatial resolution and accuracy to navigate tight gaps in city traffic and cover potential blind spots on hilly terrain.

The combined inputs from all these units overlap. Machine-learning algorithms tested first on simulators in the lab and then on public roads allow the Waymo Driver to navigate the complexities of city driving and highway driving.

In city driving, the short-range data is crucial. On the highway, long-range vision is also vital. The peripheral vision reduces blind spots caused by parked cars, or large vehicles in proximity. For example, the peripheral cameras can look past a truck in front of the car, allowing the Waymo Driver to decide if it’s safe to overtake.

Waymo claims its own LIDARs can view objects both close up, and more than 300 meters away. The best off-the-shelf LIDARs only have ranges of 250 metres or so. In addition, this Waymo system is robust enough to handle poor weather, rain, snow, hailstorms, bad roads, etc.

There’s a long way to go however, before most jurisdictions will allow fully-autonomous cars to operate, especially in crowded urban environments. However, long-haul driverless trucks are already operating in the EU and the US.

It may be argued that driverless cars will be safer than the vast percentage of human drivers. An AI system will not jump traffic lights, or indulge in other risky behaviour and most accidents are caused driver idiocy. If every car on the road is driverless and they all talk to each other, accident rates will drop sharply.

But driverless cars do make errors — there have been fatal accidents involving driverless vehicles and there are the associated issues of liability. If there’s an accident involving a driverless vehicle, who is responsible? Insurers would really like answers to that question.

There’s the other question — the so-called trolley problem, which might make some owners queasy: Let’s say the car has the option of driving off a mountain road, or hitting a minivan full of children: How would the vehicle be programmed to act? Would it opt to kill the occupant in preference to killing the children?
 

Regardless of regulations, all modern vehicles will become more and more autonomous. Less elaborate systems will be adapted for use in “non-autonomous” vehicles but those will also increase the vehicle’s autonomy quotient. Also as economies of scale take hold, these new cameras and LIDARs sensors, will undoubtedly find use in other industries.