New Yorkers Want Driverless Cars to Prioritize Pedestrian Safety

Photo: Foo Conner/Flickr
Photo: Foo Conner/Flickr

The vast majority of New Yorkers want vehicle manufacturers to prioritize pedestrian safety in any self-driving cars deployed on city streets, according to a poll commissioned by Transportation Alternatives.

Conducted by Penn Schoen Berland Research, the poll presented 880 likely New York City voters with a version of the “trolley problem” thought exercise.

“In the poll question,” says TA, “a self-driving car has only two options: swerve to avoid hitting a pedestrian, which may cause the car occupant to be injured, OR continue and hit the pedestrian.”

In that scenario, Mercedes-Benz has said its self-driving models will prioritize vehicle occupants.

The sample of likely voters skewed toward car owners (68 percent of respondents) compared to the overall NYC population (46 percent of households). Nevertheless, 80 percent of respondents said the car should swerve to avoid the pedestrian, while 5 percent said the car should prioritize the safety of the vehicle occupant.

“New York will play a leadership role in the regulation of autonomous vehicles, and this poll sends a powerful message that protecting the most vulnerable people on our streets must be the priority for cities managing these driverless cars,” said TA Executive Director Paul Steely White in a statement.

Policy makers and advocates have been developing frameworks for managing self-driving cars in case the technology matures to the point where it can be brought to market. The National Association of City Transportation Officials last year released a position statement saying self-driving cars must enhance safety for people outside vehicles and operate in a way that “rebalances” street space in favor of active transportation and transit.

The technology isn’t ready for city streets yet. Last month Uber deployed self-driving cars with human “back-up drivers” in San Francisco, in violation of California law, and was forced to retreat. One problem with the cars was their inability to properly navigate bike lanes.

TA says it will release more results from the poll over the coming weeks.

  • AMH

    I reject false choices like the “trolley problem”–the solution is to require the car to operate at a safe speed that does not require killing anyone inside or outside the vehicle. To do otherwise is incredibly irresponsible.

  • Jeffrey Baker

    Indeed, the “trolley problem” is a non-problem for autonomous vehicles. Autonomous vehicles will never be in this situation in the first place.

  • Larry Littlefield

    So you hope. But that assumes they are all autonomous vehicles.

    A truck, bus or SUV runs a red light at 40 mph. Driver error, program error, mechanical failure, hacking, terrorist, whatever.

    A smaller, self-driving motor vehicle has a choice of slamming into it or driving into pedestrians. I assume it would apply the breaks in either case.

    Or, a child runs out in the street. Should the motor vehicle smash into parked cars, causing expensive damage, or go straight on through the child?

    How about a dog?

    Again, I assume the brakes would be applied in either case.

  • Jeffrey Baker

    People don’t realize how much sensor inputs these cars have. Autonomous cars are not taken by surprise by trucks running red lights because they see the truck way before a person would have. Same with the kid: the car is tracking the kid from 100 yards off. Most of the time when a meat-driver says “that kid came out of nowhere” they really meant “I was staring at Snapchat and didn’t see the kid”. Autonomous cars never make that mistake.

    If you look at all the fatalities on NYC streets last year, none of them were of the suddenly-appearing-child variety.

  • Even if the trolley problem doesn’t illustrate a likely real-world scenario, for a survey like this it’s a useful shorthand to gauge people’s priorities.

  • Larry Littlefield

    “Autonomous cars are not taken by surprise by trucks running red lights because they see the truck way before a person would have. Same with the kid: the car is tracking the kid from 100 yards off.”

    You can’t be writing from NYC. The kid is running out from between parked cars, perhaps parked SUVs. The truck pops out from behind buildings at the curb line.,-73.9909077,3a,75y,308.93h,70.52t/data=!3m7!1e1!3m5!1sU6ASjq39nmrd4L8ecZYVpg!2e0!!7i13312!8i6656

  • Jeffrey Baker

    Children are not created by spontaneous generation in the spaces between parked cars. That image you posted is actually a good example because it’s at approximately the height of the lidar unit on Google’s cars. The car tracks people on the sidewalk and when one of those people just disappears, the car doesn’t just thing “aw fuckit, another person disappears”. It is programmed to infer that losing track of a person means to slow down. And an autonomous car wouldn’t go more than 20MPH in that situation anyway.

    Most people have a hard time internalizing this, but autonomous cars drive very carefully, have long-range vision, never go too fast for conditions, and react within milliseconds. They are rarely or never going to encounter the situations that meat-drivers believe are difficult.

  • walks bikes drives

    How about the car that pulls away from the curb or the double parked car that pulls away without warning? Autonomous car sees parked or stopped car and goes around it as it must, but driver then moves forward. It might be a robot, prone to fewer “distractions,” but not everything is predictable, even to an algorithm.

  • walks bikes drives

    So it will travel while stopped? There is no way a moving vehicle, or moving pedestrian for that matter, can travel at a speed where anyone they bump into is guaranteed to survive. Example, the elderly man who slipped on the ice in Today’s Headlines. Could have bumped into someone else while walking instead of slipping on ice. We need to minimize issues, and travel speeds of 20mph on most crosstown streets and 25mph on most arterials would be a great start. But you can’t prevent EVERYTHING.

  • walks bikes drives

    If the car is coming down 97th street and has the light, and a truck is barreling up Madison Ave in the right lane at 40MPH, unless the autonomous car can see around corners, it is likely going to get crushed. Your example holds true in wide open spaces, but not in a city.

  • Jeffrey Baker

    The car is coming down E 97th, with a speed of 20MPH from which it can stop in 25ft with a reaction time of zero. Autonomous cars can in fact see around corners because they have sensors at the very front of the car, which is an major advantage over ape-driven cars where the ape sits amidships.

    When an autonomous car can in fact not see around corners, it slows down. Simple as that. People don’t, and they get into wrecks.

    People see a truck barreling down on them and they slam on the gas, jerk the wheel, and plow into a bus stop full of little kids. Computers just hit the brakes. No moral dilemma anywhere to be found.

    The entire genre of self-driving car moral dilemma is the last gasp of the human driver who can’t imagine being replaced. It’s a stupid mental exercise suitable only for stupid people. Tens of thousands of pedestrians are hit by cars every year. Practically none of these occur in moral dilemma territory. Distraction and driving too fast for conditions are the universal causes of pedestrian casualty.

  • AnoNYC

    Imagine a city with sensors that detect every vehicle and share that information with them. Combined with radar and lidar, even a collision like that is unlikely.

  • AnoNYC

    Cars must, and will eventually communicate with one another and city infrastructure.

  • Larry Littlefield

    There are other questions. How fast would driverless cars be allowed to go when making turns? As fast as the car’s handling allows, or so slow then could stop on a dime?

    A driverless car is traveling at 25 mph and approaches an intersection with limited visibility (all of them in central areas of NYC). Would it automatically slow to 15 mph?

    A driverless car pulls up alongside a bicycle while planning to make a turn. Would it slow and pull in behind the bicycle to avoid a right or left hook, or try to get in front?

    Would NY drivers want a driverless car that made all the conservative choices?

  • Brad Aaron

    I would have agreed with you until Uber used a major city and its unwitting residents as a robot car laboratory and dared authorities to do anything about it.

  • Jeffrey Baker

    That’s true, I don’t include the death throes of the anarcho-libertarians over at Uber. They desperately need to appear to have autonomous driving technology before they run out of money. They are 5 years behind the competition.

  • Joe R.

    Your scenario merely illustrates why it’s a bad idea to mix autonomous and human-driven cars. Autonomous vehicles would communicate their location and intent to other nearby autonomous vehicles all the time. The can effectively see around corners. Allowing human driving once autonomous vehicles prove themselves would throw a monkey wrench into the works.

    It remains to be seen how well autonomous vehicles will be able to detect people. My understanding is they would scan for life signs, probably detecting people long before they’re visible.

  • Joe R.

    Also, the car will probably have multiple means of detecting people, even when they’re not visible, like heart/respiration sounds, foot steps, biological electric fields, perhaps novel ways not yet invented (think Star Trek when they’re scanning for “life signs”).

  • walks bikes drives

    Do you really think an autonomous car could hear someone breathing, their hear beating, or their footsteps anywhere in Manhattan?

  • walks bikes drives

    Simple: require all cars to be equipped with a transponder. Now, even if a car is being driven by a human, say, a ’57 Chevy, an autonomous car can have a real-time view of where everything is. It will know a car is pulling out of a parking space even when a wall is blocking it.

  • Joe R.

    Today, no. In five years quite possibly. I’m thinking of something similar in principal to telescopes with adaptive optics which compensate for atmospheric distortion. In this case call it adaptive acoustics which compensate for background noise. Also, Manhattan isn’t necessarily a place where longer range pedestrian detection matters. It matters more in the outer boroughs where an AV might be traveling up to 40 or 50 mph if it detects no obstacles. It’ll need to detect potential conflicts far enough in advance to give it time to slow down before reaching the objects. In Manhattan with pedestrians everywhere we’ll probably limit AVs to 15 or 20 mph all the time so they’ll be able to stop very quickly before hitting anybody. Hence, no need to detect a person from a block or more out.

  • Joe R.

    You could do that but the transponder can’t signal the intent of the human driver. That will mean AVs will need to be programmed to effectively yield all the time to human-driven vehicles just as a precaution. A second problem with this approach is you can’t get rid of traffic signals or stop signs. You still need them for the human drivers. Without human drivers AVs don’t need traffic signals. They can avoid each other. They can be programmed to always yield right-of-way to pedestrians or cyclists.

    No good reason we should continue to allow human driving once AVs prove themselves. Two or three generations from now kids will actually wonder in disbelief that we actually let humans drive multiton conveyances with hundreds of HP in close proximity to unprotected humans.

  • Tom B.

    The priority should account for the velocity at impact as well as the safety features for the occupant and choose the option with the lowest probability of causing any fatality without prioritizing either party’s life. Otherwise I would hope in the first death where a different choice would have likely reduced the the probability of fatality, the maker of the software would be charged with manslaughter.

  • reasonableexplanation

    Lots of good, basic stuff comes out of considering the trolley problem.

    For example, in a scenario where the only choices are to kill the occupant of the car or to kill a pedestrian, would the desired outcome be affected if the pedestrian was crossing against the light vs crossing with the green?

    What about if there is one person in the car vs two people in the crosswalk?

    However, realistically, self driving cars are unlikely to be programmed to solve the trolley problem in any meaningful way, more likely they’ll just be programmed to stop as quickly as possible when an obstacle is encountered, and for the vast majority of cases, this will be more than enough.


Informed of Safety Benefits, Most NYC Voters Want Protected Bike Lanes

A poll released by Transportation Alternatives today [PDF] sheds some new light on how NYC voters feel about street redesigns and automated enforcement when the policies are framed in terms of safety benefits. Opinion polls by Marist, Quinnipiac, and the New York Times have consistently shown that New Yorkers support bike lanes by a large margin. […]

NYC’s Most Frequent Voters Depend on Transit, Want Safer Streets

Despite the millions of New Yorkers who ride subways and buses and the thousands of life-altering traffic injuries suffered by residents of the five boroughs every year, transit and safe streets are typically second- or third-tier issues in the city’s big-time political campaigns. The 2013 mayoral election could be different: NYC DOT’s program to re-orient […]
Will the three men in a room join New Yorkers who support slowing drivers near schools?

Most New Yorkers Who Own Cars Support Speed Cameras

Right now, Albany limits NYC to 140 speed enforcement cameras for all 6,000 miles of surface streets in the city. A broad spectrum of New York City voters approve of expanding the program. The question is whether Albany Democrats Andrew Cuomo, Jeff Klein, and Carl Heastie will take action to save lives.