Back in the 90’s, when I was still a child, I was convinced that by the year 2000, cars would be able to fly. At that time, I was unaware of aerodynamics, and remember asking my dad if his Volvo 440 would take off if we opened the front doors. Dad laughed kindly and replied “If only it was that simple!”.
Now we are in 2017 and, to my dismay, flying cars have still not replaced regular ones. However, since then, the kid in the back of his dad’s car has become an engineer, obtained a PhD and is lecturing in computer vision and autonomous systems at Cranfield University. Today I contribute to the next exciting challenge and the future of transport systems: driverless cars.
Launched in 2009, the commonly called ‘Google cars’ from Google’s self-driving car project, (now Waymo) have shown ever-increasing achievements towards autonomy in vehicles driving on public roads.
This has certainly triggered today’s craze for autonomous driving. It is now not only car makers, but also tech companies, academics and industry leaders who are racing to win this technological quest. Another important factor is the emergence of ‘off-the-shelf’, affordable parallel computing hardware. Also known as a ‘GPUs’ (or Graphical Process Units), these allow the computation of a huge number of operations at the same time, very fast. This combines with an increasing amount of available data, enabling the resurgence and practical expansion of artificial intelligence (AI).
Humans learn mainly through observation, and our eyes are one of our principal sources of information, feeding into our brain. Similarly, digital imaging sensors provide a huge amount of information to computers. One way to teach autonomous cars how to stay in lane for example, is to ‘train’ them with images from a camera showing the car’s front facing view. This is matched with other inputs recorded by human drivers, such as the angle of the steering wheel, car speed, the levels of both the accelerator and brake pads. The machine then uses AI to model the human driver’s behaviour and create rules for driving. For example, when pedestrians are crossing, the car brakes; when traffic lights turn green, it accelerates; and when it sees curved lines on the road, it turns.
As well as reducing the stress of driving, autonomous vehicles might offer much more in terms of social and environmental impact. A transport system where autonomous vehicles become a shared commodity, available to anyone at any time, could solve a number of issues. Through the vehicles’ association to big data and inter-connectivity, they would enable congestion-free traffic, ease the strain on parking, reduce pollution and increase safety.
Alongside the technical challenge of autonomy, self-driving cars face many social issues before becoming an ideal transport system. These include issues relating to regulation, the transition from ‘driven’ to ‘driverless’, protecting against hacking, and the complicated ethics of removing the driver. However, this process is underway and is bringing with it many opportunities for new engineers to test their ideas.
Lounis is passionate about extracting and interpreting all kinds of information that can be found in an image. As he likes to say, he loves to “make the images talk”. He also, has a great interest in education of youth especially following the ‘Teach Your Own’ methodology and cares about environment with a particular interest in permaculture. He hopes one day to concentrate his efforts to combine education, respect of nature and technology, for the benefice of people.
Latest posts by Dr Lounis Chermak (see all)