Lidar for riders on the autonomous car storm: Lidar and computer vision systems from 1995 to today

The story starts in 1995. A silver Mercedes Benz S-Class W140 drags past more conservative drivers on the German Autobahn at 110 miles per hour (180km/h). It might sound like nothing out of the ordinary. Except this car had a cool name: VaMP. And it was autonomous.

VaMP was an impressive early example of an autonomous car. Adding to the top-trump speeds were the 1,043 miles that it drove from Munich to Copenhagen, and the commendable distance between human takeovers: the longest uninterrupted distance was 158km, with a mean of 9km. This is a better average “disengagement rate” than Baidu’s July 2017 average of once every three miles. Waymo, the Alphabet company, admittedly did better with an average of once every 2700 miles in April 2017.

Preparing for a long-distance drive in the 90s

The self-driving Mercedes was the result of an eight-year-long European project named PROMETHEUS that aimed to develop systems capable of autonomous mobility. The car used a computer vision system that processed real-time images from four cameras with two different focal lengths for each half of the car. The cameras relied on artificial controlled saccadic movements – similar to the way that the eyes jump from object to object and retain focus – to focus attention where it was needed. The information from the cameras was fed into sixty transputers, a type of parallel computer, that in turn fed instructions to the steering wheel, throttle, and brakes.  

Also, in 1995 a team from Carnegie Mellon University in the USA drove a modified Pontiac minivan that could control a steering wheel and stay in lane across the USA. This was also using computer vision to map out the road.


The lack of publicly visible progress in self-driving cars between 1995 and 2009 (when Google announced it was working on an autonomous car) would appear to show that the crucial computing, mapping and vision systems had not reached a satisfactory level of development. One thing that most companies involved in autonomous cars agree is that lidar (a visible-light version of radar) in tandem with computer vision and other sensing systems is a crucial component of this new wave of autonomous vehicles. This article will focus on the impact of lidar. Other future articles will consider other advances such as processing power, wireless connectivity and machine vision.

In 2005 Velodyne produced a novel lidar rig targeting the autonomous car market. It had 64 lasers on a rotating gimbal that could build a 360° 3D image of its surroundings. This electro-mechanical type of lidar would become the centrepiece of the iconic Google Firefly self-driving vehicle. Lidar’s advantage over systems such as the VaMP’s that solely used cameras to see the world is that it is much better able to build up a 3D image of its surroundings producing more accurate perception of depth and objects. This is important for cars that will be in complex inner-city environments where cars need to distinguish young children from a leaf of the broadsheet newspaper tumbling across the road propelled by the wind. However, electromechanical lidar is bulky, expensive and has moving parts which quickly wear out. This may not seem to be a problem but at USD70,000 per unit it negatively affects the price of an autonomous car for the ordinary person.

Another type of lidar has been developed recently: solid state lidar, which promises cheaper price points, smaller size and lengthened life-time. One type is the phased array lidar that uses the same principle as solid-state radar: sending out signals from a phased array of lasers and measuring the light’s bounce-back time-of-flight to build a “point-cloud” image of an environment. The model is the primary input to the vehicle’s autonomous control system. These solid-state systems have no moving parts, which increases the lifetime to 100,000 miles before they need replacing; electromechanical lidar systems require replacement at a fifth of the distance. Quanergy is producing solid-state phased-array units for a retail price of several hundred dollars to a few hundred depending on quantity, with aims to reduce that to USD100 soon.

With a Waymo autonomous Toyota Prius estimated to currently cost USD320,000 that’s a big deal. These considerations are important for the future of autonomous cars as they can determine the ownership model; eg whether autonomous cars are owned exclusively on a fleet model, or private ownership as today.

[Image licensed to Ingram Image.]

Add this: