The world is still several years away from realising a driverless future, despite promising developments and new capabilities in self-driving cars.
Progress has been made in incorporating LiDAR technology and creating maps specifically for artificial intelligence (AI). In the United States, the National Highway Traffic Safety Association is contemplating revisions to vehicle safety rules to better accommodate self-driving cars.
However, building a fully capable self-driving car has hit a couple of speed bumps along the way. What are some of the challenges that are hampering progress?
When it comes to self-driving cars, the top priority that comes to mind is ensuring safety. While self-driving systems have made good advancements in detecting pedestrians, signs, and other cars, there’s still a long way to go.
At the moment, it is rare for vision systems in autonomous cars to consistently reach over 99.99% accuracy. According to Amnon Shashua, CEO of Mobileye, the best systems often incorrectly perceive the surrounding environment once every thousands of hours. That could make accidents uncomfortably frequent.
One way to get over this hump is to integrate several vision or perception systems, multiple backups, if you will. A combination of cameras, radar, intricately detailed maps, and other systems can make it easy for a car to perceive its surroundings or a specific situation.
When all these different variables are detected, just how reliable are autonomous vehicles in making the right decisions?
Driving involves managing a multitude of interactions while moving along the road or highway.
Humans are significantly more capable of managing a variety of driving-related activities and coming to a reasonable decision on a course of action. Whether it’s giving right of way to another driver, ensuring the bicycle rider has enough road space or signalling pedestrians to cross.
On the other hand, a robot doesn’t have the benefit of human common sense and judgement. Currently, self-driving cars still cannot effectively differentiate between a person standing in place and a person who looks likely to cross the street absent-mindedly.
Some companies developing driverless technology, like BMW and Audi, are sticking to the quick and easy solution of integrating both human and computer control. This allows the latter to hand back the driving wheel to a human in tricky situations.
Others like Google are more ambitious in pursuing a completely driverless set-up, but for that to be achieved, autonomous cars must improve their understanding of why pedestrians or drivers are behaving the way they are in any given scenario.
When a sophisticated computer system is essentially responsible for the safety of passengers, one can’t help but think of how serious a threat hacking may pose. After all, autonomous vehicles are basically computers and if a hacker can infiltrate such a system, what’s stopping them from causing unnecessary casualties?
Experts foresee traditional car companies will initially struggle to address this. The car industry has never had to worry about cybersecurity issues until now. However, tech companies like Google, Uber and Lyft might have a slight advantage as these are organisations that have a better grasp of managing data and building AI systems.
Overall, both car and technology industries would need to continuously adapt to ever-evolving security threats.
Both the public and private sector would need to adapt
A driverless future may be on the horizon but different parts of society would need to adapt to that reality. Regulatory policies would need to amend the definition of a driver and online fleet management software would have to develop systems to manage complex computer systems within vehicles.
While overcoming the challenges mentioned above will get us considerably closer to a driverless future, a different set of obstacles await autonomous cars.