Smart infrastructure may be the cheapest driverless option

For all the talk of LiDAR, radar and connectivity in self-driving vehicles, little has been said regarding how the expense of these features could influence widespread adoption. Will carmakers favour one over another if it’s cheaper to produce and, if so, how will this impact when and where the cars are deployed?

Edwin Olson, CEO of May Mobility, is thinking about this dilemma. His start-up has already deployed its own infrastructure (with DSRC-like capabilities) to support the autonomous shuttles it builds in Ann Arbor, Michigan. This has increased reliability and reduced the cost of each individual shuttle. However, Olson admits the company may not need that infrastructure forever.

“But it may still be a cost advantage for us to use it because it might decrease the cost of the individual vehicles,” he said. “For our deployment it makes a lot of sense because we’re not trying to take on a whole city at once. If you had to put a computer on every traffic light in an entire city, that’d be a significant up-front capital expense. We’re looking at more targeted routes, where we know there are 10 traffic lights and we’re going to use those lights many, many times. The economics of deploying technology in those environments is pretty favourable.”

Olson said it is “absolutely possible” that a connected infrastructure will prove to be more economical than a pricey sensor array. If so, the auto industry is likely to rely on V2I even if it isn’t necessary.

Who will pay?

Electric carmakers have made the unique decision to pay for many of the charging stations themselves. Now May Mobility has set an interesting precedent by paying for its own infrastructure. Does that mean others will do the same?

Said Olson: “I think for things that require cooperation with traffic lights, like a more pervasive deployment of DSRC, an external entity is not really going to be able to come in and plug into the traffic light system. So, there are some roles that are best served by the municipality but for other kinds of sensors that might have a shorter lifespan, I think it’s reasonable to expect the transportation providers to pay the bill for that themselves.”

The complexities of freeways

It’s easy to underestimate the subtle complexities of freeway driving, which may appear to be less chaotic than traditional roads. That is true in some respects but Olson noted that freeways are also an area where tiny mistakes can turn into serious problems.

He explained: “Freeways are kind of an enigma, in a way, for autonomous driving. On the one hand it’s much more structured than your typical urban environment. You don’t have skateboarders going across your path. Everyone is going pretty much the same direction. The speed between cars is relatively low.”

The challenge, however, is avoiding those small, seemingly insignificant mistakes. “That means that you have to be very sensitive to making very small errors,” Olson added. “That’s a real challenge. We don’t think the technology exists to do that safely at a level necessary to put these into highway deployment.”

Building smarter cars

Google was forced to make its own vehicles drive more aggressively after it learned that the previous approach, in which the cars were too cautious, might have been responsible for causing some accidents. “There are a lot of tricky situations that a car can find itself in,” said Olson. “The approach that most people take is to build something that’s pretty conservative and, if something is moving, that car will then slam on its brakes.”

Olson said that kind of behaviour is “its own source of danger”. He added: “The challenge is to understand well enough what’s going on around you to realise that the pedestrian around the corner isn’t someone you should be slamming on the brakes for. That person is going to continue minding its own business.”

Not even human drivers are capable of reading the signals (or lack thereof) from pedestrians, so it could be very difficult for AV makers to perfect this element. “Imagine a human driving down the road and you see a pedestrian standing on the side, holding their cell phone,” said Olson. “Now what you do as a driver largely depends on if their face is down at their phone or up and looking at you, the oncoming car. In the former case you’re likely to slow down, in the latter they see you and you develop that sense of trust.”

In the end, Olson thinks that both AVs and human drivers need to understand the subtle cues of pedestrians. Regarding AVs specifically, he said: “It really comes down to giving the system as much data as possible so it can make the best predictions that it can.”


Leave a comment

Your email address will not be published. Required fields are marked *