If people outside of the car industry have heard of lidar, there’s a good chance it’s thanks to Tesla CEO Elon Musk. While most teams working on self-driving cars see the light-based technology as an essential tool, Musk insists it’s a false hope – an expensive misstep that distracts from perfect, camera-based vision. Musk and his competitors hold competing visions for the route to autonomous vehicles. But is the famous entrepreneur correct?
Elon Musk has made no secret of his disdain for lidar. As CEO of Tesla, Musk is doubtless the single most recognizable figurehead for driverless cars, yet he’s frequently been disparaging about a technology that his rivals see as a key tool. No Tesla car has lidar and, at Tesla’s 2019 Autonomy Day, Musk expanded on why. “Lidar is a fool’s errand,” he said. “Anyone relying on lidar is doomed!”
To understand the strength of Musk’s feeling, and why others are sure he’s wrong, we first need to understand the ways that autonomous vehicles ‘see’ their way in the world. Increasingly, today’s level one and two autonomous cars are reliant on a mix of camera and radar sensors. While the former is used to recognize road signs or help eliminate blind spots, the second is needed to measure distance, direction and speed of other vehicles or obstacles.
While most new cars now have some idea of the environment around them, this needs to be dramatically improved before fully-autonomous vehicles are ready for the world’s roads. Richer data such as high-resolution mapping may well be key to cars advancing to level three, where they’re capable of driving themselves in limited circumstances. However, to reach full automation, vehicles will need to understand and react to the dynamic environment around them, just like a human driver does.
Seeing the Light
On one level, light detection and ranging (lidar) is simply another technology for vehicles to view the world around them. Similar in concept to radar, lidar sensors fire pulses of invisible laser light in an arc around the vehicle, measuring the round-trip time of the reflection. Coupled with powerful computer processing, this lets the vehicle build up a detailed and very accurate 3D view of the world around it.
But while radar sensors are in mass production and are affordable, lidar for vehicles remains in development. Lidar sensors are also comparatively expensive and potentially too fragile to withstand a lifetime on the road. This is the practical reason why no Tesla car is sold fitted with lidar – the tech is simply not considered production ready. So why is lidar fundamental to every other major self-driving car project?
The simple answer is that, as good as computer vision currently is, systems based on cameras and radar are still some way short of matching humans when it comes to judging distance and avoiding hazards on the road. Teslas are in fact a great example of where the tech currently stands – some way off achieving the human-like levels of visual understanding necessary for full autonomy.
Elon Musk understands that Tesla can crack the vision problem with a combination of processing power and extensive machine learning, being in the unique position of having all the data gathered from the thousands of Tesla vehicles already in daily use. Tesla’s competitors don’t have access to the same volume of real-world driving data, which means that they forced to find other ways to accelerate their path to self-driving.
It’s here that lidar and radar come into their own. Onboard computers aren’t yet reliable enough when calculating the distance and relative path of objects picked up by the vehicle’s cameras, but radar and lidar information makes that possible. While radar has the advantage that it can work in almost any weather, lidar produces the most accurate measurements and, potentially, has a higher resolution.
By designing vision systems that combine camera, radar and lidar input, companies like Waymo probably don’t need to fully solve the computer vision problem before getting a self-driving car on the road. Instead, they’ll rely on ‘sensor fusion’ – the knitting together of camera, radar and lidar data to produce a coherent and comprehensive view of the environment around the vehicle.
Ultimately, Musk is betting that Tesla will crack the computer vision puzzle using only cameras and radar, making lidar redundant. In the meantime, his competitors are taking a shortcut by combining cameras and radar with lidar, a tactic which they hope will be cheap and robust enough to include in production vehicles once their self-driving technology is proven. Whoever gets to market first will have a huge commercial advantage, which partly explains Musk’s outspoken defence of Tesla’s approach.
For the rest of us, the key point is that both approaches are likely to succeed. This is why VIA is working with partners across the sensor space, and supporting sensor fusion across multiple technologies, crucially so that solutions can be tailored to the specific needs of a customer. Whichever method proves most successful, the end destination is the same – fully self-driving cars within the next 10-20 years.
At VIA, our mission is to enable companies to accelerate the development and deployment of innovative new transportation solutions and services that redefine safety, convenience, and comfort to give our customers a competitive edge as we enter a new age of seamless mobility and autonomous driving. To learn more about VIA Mobile360 solutions, click here