Tag: av

Designing AVs for people

Rakesh AgrawalLeave a comment
Elderly woman with cane standing on sidewalk near crosswalk as autonomous car approaches stop sign

My dad had to walk an extra 1/3 mile the other day. With a cane.

Our Waymo zipped right in front of our destination, circled and then came to a stop a few blocks away.

As is often the case in technology, the designers didn’t have any empathy.

They optimized based on moving machines around, not moving people around. Those are fundamentally different optimization problems. Unlike products at Amazon fulfillment centers, people aren’t payloads. (Most of this post is focused on truly autonomous vehicles; I’ll touch on L2+ like the Mercedes Benz/Nvidia solution later.)

For an able-bodied person in good weather, walking 1/3 of a mile isn’t a big deal. For someone who needs a wheelchair or a cane, it is a very big deal. Others don’t want to step into snow and slush if it can be avoided.

Ride apps like Uber and Lyft have a big advantage: drivers can localize for the conditions based on what they see. A human driver isn’t going to drop you at the bottom of California St. with a 25% grade. (That’s not even the worst in SF; one street goes to 41%.)

In my dad’s case, we would have asked the human driver to stop right in front of the Ferry Building. He’d turn on his flashers, double park or inch his way into the right lane to make that happen.

Waymo has another problem: PR and regulators. Because it’s vehicles are a driving billboard, mistakes like double parking will be posted on social media. Regulators will see the most egregious ones and hear about them.

On the other hand, disability activists will protest that they are being dropped off in places where the service is not useful. That again will get to regulators by way of interest groups.

When I design software products, unnecessary friction means an extra click or a confusing interface. I optimize as many of those out as I can. But when I develop hardware, it’s a whole different problem and the shipping bar is much higher.

Driving is a shared environment. We’ve long had bicyclists, motorcycles, pedestrians and cars sharing the road, curbs and sidewalks. Recently, we’ve added Lyft/Lime scooters. We’re now adding delivery robots and AVs.

In order for this to work in an environment like San Francisco and New York, we have to tolerate things that are technically illegal. If we didn’t, SF and NY would be paralyzed. This is so well accepted that UPS and FedEx get special high-volume rates for their double parking tickets.

In this video, Nvidia CEO Jensen Huang comments as his AV Mercedes drives through San Francisco. Double-parked car? Just go around through on-coming traffic. Technically, illegal, but realistically necessary.

Mercedes and Nvidia have the advantage that their vehicles aren’t driving billboards. (There are a view test vehicles that have Mercedes and Nvidia test vehicle markings.) Like Uber and Lyft, they have the benefit of adaptability by humans for the places where it is needed. Also like Uber and Lyft, they are driving in incognito mode.

Whether it is L2+. L3 or higher, we’re going to need to re-negotiate our social contracts. AVs will have to adapt, as will humans.

Redesigning the self-driving future

Rakesh Agrawal

I am driving with Tesla’s Full Self-Driving (Supervised). As I approach, the steering wheel seems to be freaking out, moving back-and-forth quickly. The car’s apparent freak out causes me to freak out.

After a few turns like this, I realize what’s happening: the car is making micro adjustments during the turn.

A human doesn’t make those micro adjustments. We wouldn’t be capable of fine tuning that rapidly. We don’t have to be perfect, we just need to get where we are going. If we’re a little bit off the center arc of our turn, that’s fine.

It would have been a more comfortable experience if the car just smoothed out those micro adjustments. They would still be done, but anything that was tiny wouldn’t provide feedback via the steering wheel.

This is similar to when anti-lock brakes came out on cars. The long-time advice in driver’s ed class was that you should slowly pulse the brakes to avoid going into a skid. With ABS, it is different. You should stomp on the brakes. The ABS then pulses the brakes faster and with more control than a user could. As with the Tesla steering wheel, the brake pedal provided vibrating feedback – not a comforting user experience.

I was in a Waymo the other day. Because the ride was so comfortable, I sat back and enjoyed the conversation with my friend. That was interrupted by a “click-click, click-click, click-click.” The car was indicating that it was about to make a turn. The signal action is very useful information for the cars and pedestrians around it. The clicking is useful for human drivers. It isn’t useful or necessary for passengers in a self-driving car. (In the transition period, the clicking might make passengers more comfortable. Perhaps it could be a preference setting.)

We are going through a major transition in transportation. As we design the new experiences, we need to be cognizant of:

  • What makes drivers more comfortable in a semi-autonomous environment? (L2/L3)
  • What makes passengers comfortable in an autonomous environment? (L4/L5)
  • What is vestigial that we can remove to create a better experience for both? (L3-L5)
  • How does the vehicle interact with the overall environment? (L3-L5)

We need to be looking at what we need to do during the transition as well as in the ideal future state.

You could have electrochroamatic glass that provides privacy to occupants. If the car is self-driving, you don’t really need to look out, do you?

One of my really future state ideas is auto-off headlights in autonomous environments. It’s the opposite of auto high-beams. When there is no one or no moving objects nearby, turn off the headlights and noisemakers to reduce light and sound pollution.

Of course all of this involves navigating a complex regulatory landscape that can vary by city, state and country. For the U.S., I’d like to see federal preemption. That might take a while. In California, cities like San Francisco, don’t even want state preemption.