Teaching Autonomous Vehicles To Drive With Common Sense

The journey towards seeing autonomous cars on our roads in any meaningful way has perhaps been a longer one than many would have hoped for, but with human error contributing to around 90% of the 5 million accidents in the United States per year, it’s a mission that still has undeniable benefits.

Research from Harvard Business School suggests that adopting autonomous vehicles en masse would have a similar impact on global health as the introduction of vaccines.  The rollout has been slow, however, not least due to the resistance among people to relinquish control of their vehicle.

Human driving

Central to many discussions on the technology has been the trolley dilemma beloved of ethicists, but the authors believe these kinds of no-win scenarios are the wrong way to look at the issue.  Instead, they argue that the best way to get us to trust the technology is to imbue it with the kind of common sense human drivers themselves use.

“People want to be able to look at a vehicle and say, it’s behaving in a way that coheres with my understanding of what common sense driving looks like,” the researchers say.

This means that the question is not so much the extreme scenarios described in the trolley dilemma and more the run-of-the-mill scenarios in which human drivers so often bend the rules of the road.

“Human drivers take a test, and we then assume that they are going to use our shared human psychology to reasonably deal with what they encounter beyond the test,” the researchers continue. “But we can’t just make the same assumptions about AVs because common sense is part of what’s being engineered into their systems in the first place.”

What is expected

To expect autonomous vehicles to drive with common sense may depend on them understanding what we tend to expect of other road users.  For instance, if an AV slows near a pedestrian crossing, pedestrians may believe the vehicle has seen them and is giving way to them, even if it hasn’t in fact done so.  Sending more explicit signals as to the AV’s intended behavior may be required to ensure it is safe.

It may ultimately require a complete refashioning of our driving systems.  This could enable cars to be trained to handle a wide range of different scenarios.  The authors believe that a key standard to aim for is something they refer to as SPRUCE, which suggests we should teach vehicles in a way that is:

  • Safe—“does not harm others or put others at unreasonable risk of harm;”
  • Predictable—”AV’s maneuvers can be anticipated from past behavior;”
  • Reasonable—”does not offend notions of logic or justice;”
  • Uniform—“treats seemingly like situations alike;”
  • Comfortable—“physically and psychologically smooth;” and
  • Explainable—“fits in an accessible narrative of cause and effect.”

“We also shouldn’t assume there’s always a binary right or wrong answer,” the researchers explain. “It’s more like a hierarchy of preferences. For instance, you might put safety first, following road rules second, and helping the flow of traffic third. But you also need to know when to violate these rules and by how much in order to drive in ways that are more practical.”

Gradual progress

Suffice to say, achieving commonsense in autonomous vehicles is unlikely to be straightforward and the authors believe it will be a gradual process.  It’s likely that the vehicles will always be learning and improving via continuous software updates, while our own expectations and behaviors will evolve as we become used to seeing autonomous vehicles on the roads.

“It’s not like you can program a vehicle with an ideal set of instructions that is going to solve every problem,” the researchers say. “It’s a more gradual process of integrating into this commonsense driving world.”

It’s also likely to be crucial that manufacturers are clear and transparent with consumers about any safety challenges still faced by the technology, and indeed with what elements of it are and are not working well.

“If companies can legitimately convince consumers that they are transparent, vigilant, and continuously improving, then consumers will reasonably trust them,” the authors conclude.

Facebooktwitterredditpinterestlinkedinmail