Tech breakthroughs could provide key to safe autonomous driving
The prospect of a driverless – and accident-free – world of transportation could be closer than we think.
In a story that recently went viral on social media, the San Francisco Police Department was confronted by a novel challenge: pulling over an autonomous vehicle from Cruise, the self-driving firm backed by General Motors, with nobody in the car.
This kind of incident is only likely to become more commonplace as more autonomous vehicles hit the road, highlighting the practical difficulties of dealing with self-driving vehicles as they begin to infiltrate the mainstream market.
Although grand predictions about the “self-driving revolution” have abounded over the last few years, progress has been hampered by teething problems with the technology and a few high-profile accidents which have soured public and governmental perception.
However, new innovations are shaping up to address these issues and with the advent of complex AI-supported sensing systems and machine-learning algorithms, the prospect of a driverless – and accident-free – world of transportation could be closer than we think.
Although companies like Waymo have already been running limited robotic ride-hailing services in Phoenix, Arizona, since October 2020, the predicted expansion of such schemes into other states and cities has yet to materialize.
That’s because Phoenix’s largely flat topography, consistently fair climate and relatively sparse population lends itself well to autonomous driving, whereas the complex layout, variable weather and hectic pedestrian activity of cities like San Francisco present their own issues entirely.
Indeed, the discord between the technology’s capabilities and the real-life conditions of a busy metropolis has already resulted in numerous headline-grabbing incidents.
Tesla, for example, has been implicated in at least 12 accidents (including one fatality and seventeen injuries) through the use of its controversial Autopilot software, which, unlike Cruise and Waymo, is being beta-tested by ordinary consumers.
According to the testimony of one such unfortunate individual, his vehicle veered into oncoming traffic and resisted his attempts to correct the dangerous action.
The companies behind the vehicles have tried to explain away the accidents as being caused by human error – an argument which holds some weight. A world in which only driverless cars took to streets free from pedestrians would likely involve far fewer road traffic incidents.
However, it’s these “edge cases”, where unexpected behaviors from other road users throw the AI off-kilter, which must be accounted for going forwards.
Technological solutions to technological problems
Thankfully, there are plenty of tech experts hard at work trying to figure out innovative solutions. One such entity, Neural Propulsion Systems (NPS), believes it may have found the answer by combining multiple different sensing techniques together.
In isolation, cameras, light detection and ranging (LiDAR) and radar are insufficient for reasons of illumination at night-time, visibility in poor weather and imprecise definitions of geometric shapes and materials, respectively.
Together, however, they could compensate for the shortcomings of their counterparts to create a more holistic picture of the traffic situation.
According to NPS founder and CEO Behrooz Rezvani, the densely integrated, deep sensor-fusion system allows vehicles to see up to 1,000m ahead, including through obstacles and round corners.
This, he argues, is the key to achieving zero roadway deaths in autonomous cars, which is in itself the crucial element to widespread adoption of the technology.
“Based on principles from physics and information theory, it is possible for sensors to see well enough to enable zero roadway deaths,” Dr. Rezvani explained. “This is not wishful thinking—it’s possible today”.
Meanwhile, a team of scientists over at the Massachusetts Institute of Technology (MIT) have come up with a similarly ground-breaking means of predicting the behavior of other road users.
By assimilating masses of data on past trajectories and current map layouts, then breaking all of that information down into more manageable chunks, the machine-learning system can achieve better results and use less energy than current market-leading systems.
An accident-free world on the horizon?
If promising technologies like these can be drawn together and incorporated into existing self-driving vehicle set-ups, the long-heralded driverless revolution may finally arrive.
Given that the World Health Organization (WHO) estimates that over 1.3 million people lose their lives in road traffic incidents each year, that giant leap forward can’t come soon enough.
The limited capacity in which the tech is already being used (such as the Phoenix ride-hailing service or the Australian mining site which safeguards human life by sending robotic vehicles into dangerous locations) shows the potential – it’s now up to the manufacturers of autonomous vehicles to seize it.
That journey begins with ironing out the kinks which caused the SFPD to approach the unmanned vehicle with quizzical caution earlier this month and continues with steps to regulate questions of insurance and liability.
But it won’t end until the relevant tech is in place to make a driverless vehicle even safer than its human-occupied counterpart.
Have any thoughts on this? Let us know down below in the comments or carry the discussion over to our Twitter or Facebook.
- California’s DMV approves both Waymo and Cruise to put autonomous taxis on the roads
- An autonomous drone hunted down and engaged Libyan soldiers without any human intervention
- Police in California are using autonomous drones to track suspects
- 5th Gen Waymo driver to innovate autonomous vehicles