Are autonomous vehicles safer than humans?
We’ve come a significant way in progressing vehicle autonomy, with most manufacturers achieving level 2 of automotion out of the SAE defined 6 steps. Many are making significant inroads into level 3 with partially-automated highway driving assist systems being enabled on many cars in the upcoming year.
The complexity around full autonomy is about managing our complex road infrastructure in combination with pedestrians and people-controlled-cars. Manuel Del Castillo, GNSS expert discusses this topic and his thoughts on the challenges involved with full autonomy and if autonomous vehicles are ultimately safer than humans.
Levels of automation set forth in SAE J3016 standard (SAE International 2016). Reprinted from NHTSA (2017).
Are autonomous vehicles safer than humans?
Chances are, if you have taken a modern transatlantic flight, you've probably been flown by an autopilot from takeoff to landing the entire time. Great strides are being made in terms of vehicle technology, sensors, GPS navigation, lasers and radar. They are , however, subject to challenges of reliability, safety and cost.
Mckinsey predicts that by 2030, 12% of new passenger cars will be sold with Level 3 and above autonomy, extending to 37% by 2035. In a highly regulated space, the problem resides with the ability to train a system that can safely handle not only the rulebooks of roads and highways, but also the many exceptions, contradictions and edge cases we encounter every day.
The present level of training is still narrow
Presently, autonomous driving systems are not at all well trained in how to deal with unexpected events such as a roadside emergency.
As humans, we learn by observing and experiencing the world from a very young age. We are taught the fundamentals of maths and physics. We know about speed and distance through our own experience of the world, combined with intuition, reflexes, emotion and empathy. This allows us to deal and react appropriately in complex environments and situations. Self-driving cars it would seem, have a lot of work to do to catch up.
There's no doubt that present developments in Advanced Driver Assistance Systems (ADAS) are bridging the gap to full autonomy and will have a major role to play in improving the safety and efficiency of our roads, but complexities exist with extending these capabilities in challenging weather conditions and complex urban environments. In such situations, new technologies (including our own) are helping to pave the way to full automation.
Can autonomy replicate human instinct?
Autonomous systems, unlike humans, never get tired. Or distracted. They can take a 360 degree evaluation of their surroundings, combining a wealth of complex sensor data and have great potential in many ways:
They don’t have a limited attention span and always remain alert.
They can sense more of the world, with more sensors, in more directions, simultaneously.
Their performance is consistent - unaffected by fatigue, personality or emotion.
So we might expect that the large-scale introduction of autonomous vehicles will reduce the number of accidents on our roads considerably.
However: AVs struggle with intelligently interpreting the world, especially in situations that are unexpected or out of the ordinary. Think for example of a burning vehicle on the side of the road, or a collapsed traffic sign. Dealing with unexpected changes to our expected view of the world is something humans are excellent at. Will autonomous systems ever have the same levels of perception and adaptability that we do?
Managing the challenge of people, people-controlled-cars and autonomous vehicles
A broader challenge we face to ensure future vehicles are able to operate with less reliance on human intervention is the management of the mix of people, people-controlled-cars, and autonomous vehicles on the roads. People might think that AVs are capable of greater things than they really are - for example that they will always be able to stop on a dime whenever people step out in front of them - forgetting, or not realising that regardless of the myriad of very clever sensors and computers onboard, they are still limited by the size of their brake pads and the associated fundamental minimum stopping distances.
We need to ensure that we do not place excessive faith and confidence in the abilities of AVs, which are still at the mercy of the same laws of physics. There are always things that humans are ultimately going to be better at than autonomous vehicles - such as following social cues, reacting to the unexpected and interpreting spoken instructions given by a traffic officer.
To progress the future of autonomy, we need to move the requirement of sensing from the human driver to the vehicle. This task will include:
- Adding more sensors to cross check each other
- Improve the AI models that interpret this data
- Improve the quality of the sensors so the data is cleaner and more reliable and less likely to mislead.
So, whilst current work in this space involves investment in advanced technologies to improve and expand the capabilities of a vehicle, there is still a long way to go before AVs are able to consistently cope safely on roads where human and machines need to co-exist.
Find out how we are supporting accurate positioning for autonomous cars and driver assistance platforms here.