Chris Bessette, program manager for autonomous driving at Draper, discusses if and when self-driving cars might enter the mass market.
While the technology and testing methodology for autonomous vehicles has advanced exponentially in recent years, there are still profound challenges to overcome, largely centered around the issue of perception.
People often take for granted the remarkable effectiveness of the human eye to create high-definition images as well as the ability of our brains to process these images. Our ability to notice and understand minimal cues on the road, such as spotting a pedestrian focusing on their cell phone before crossing the road, is incredibly difficult to replicate. Even more complex is how a driver responds to these events.
There are two key ways technology works to try to mimic this ability for autonomous cars; first through sensors that scan the world around them, and then through software that interprets this information and sends a message to the car to act.
Lidar has emerged as a critical sensing modality, helping cars to navigate by providing a real-time 3D image of the vehicle’s environment. However, units frequently used in test fleets today have a price tag in the tens of thousands and their intrusive size is likely prohibitive to mass production.
At Draper, we are developing a chip-scale lidar designed for high-volume manufacturing, with a view to reducing cost closer to the hundreds per unit. This represents significant progress, but major barriers persist. Beyond the core sensory requirements, there are practical questions like how lidar operates across a variety of physical environments or how the equipment will remain clean on a dirty road.
The challenge for developers is a practically infinite design space for testing autonomous vehicles. On my journey to work in Boston, for example, I have never experienced two identical driving environments. Teaching the technology to differentiate and communicate an appropriate response to the car in any situation is therefore a profound challenge within the testing phase.
This has also shifted the industry’s methodology and approach to road safety testing. Traditionally, automotive developers have taken a miles-driven approach to testing. While this is sufficient for lower technology vehicles, it is not practical for self-driving cars. The industry has therefore pivoted toward augmenting a miles-driven approach with a simulation methodology, which offers significantly more coverage of the design space.
Simply put, driving a car at 60mph will only provide testing information at a rate of one mile per minute, whereas through simulation, developers can measure performance over many miles and through a variety of environments.
Beyond creating effective sensors at an acceptable price point, we must develop software that can absorb this information from a variety of sources, integrate it, and produce instructions that make sense for the car.
There is a widespread expectation that an autonomous vehicle will react, behave and drive in the same, if not a more effective, manner as a human being. Current driverless car systems successfully handle 90-95% of all scenarios; however the remaining 5% is proving challenging. Artificial intelligence is being leveraged heavily for this function, but again, further development is needed.
Trust also remains a barrier to adoption of driverless vehicles. In 2016, there were 34,439 fatal road crashes in the USA, but how ‘safe’ is ‘safe enough’?
If driverless cars help to reduce the number of fatal road crashes by 50%, is that sufficient? These are key philosophical questions that require tight alignment across government and industry.
Trust and adoption go hand in hand, and that may turn out to be the biggest challenge of all. While I am confident that driverless cars will enhance the safety and efficiency of our roads, more development is required to demonstrate the safety benefit across all environments.
Only when drivers are confident in the ability of technology to safely navigate our roads, will autonomous driving become a reality. We will overcome these barriers, deployment of driverless cars is not a matter of ‘if’ but ‘when’.