Once just the imagination of science fiction writers, fully autonomous vehicles are rapidly being tested around the world – but they face big hurdles before they become commercially viable for the general public. One of these hurdles is the price of LiDAR (Light Imaging Detection and Ranging) systems.
As opposed to the radar and camera solutions that are on most semi-autonomous vehicles on the market today, LiDAR (Light Imaging Detection and Ranging) uses laser light sensing to help the car develop a 3D map of its surroundings. Some experts argue that it is a safer solution, as a result.
The price-tag of some of the market-leading solutions, however — around $75,000 — has meant that fully autonomous vehicles have been out of reach to the general public. But a new player, Luminar, recently announced LiDAR-based solutions for under $1,000. The company is working with brands such as Toyota, Audi and VW on LiDAR solutions for self-driving cars.
ITU News spoke to Kevin Mak, Principal Analyst at Strategy Analytics, about what this price drop means for the future of the autonomous vehicle industry.
Most ‘driverless’ consumer cars on the road today us camera/radar solutions. What are the advantages and disadvantages of LiDAR vs. using a camera and radar systems in a self-driving car?
The advantages of these new LiDARs is the higher resolutions they can provide over existing sensors. That way, an autonomous vehicle would be able to recognise kerb stones and other road features that a camera would struggle under low light conditions.
RADARs still lack the resolutions needed for self-driving and are not as capable in detecting stationary objects. But, ultimately, an autonomous vehicle would need the use of many different types of sensors, as LiDARs are still too costly and require powerful processors to perceive objects in their Field of View.
What does this dramatic price shift mean for manufacturers?
Naturally, an autonomous vehicle using LiDARs costing in the hundreds-of-dollars is more affordable than those using LiDARs (e.g. Velodyne HDL-64) costing in the tens-of-thousands-of-dollars.
Will this development speed up the timeline for self-driving cars becoming a ubiquitous consumer product?
At Strategy Analytics, we do not see autonomous vehicles (at Society of Automotive Engineers (SAE) Levels 4 and 5) entering the market any time sooner than the mid-2020s, given the vast technical and legal challenges facing them and ensuring that they can operate reliably and safely – not just in sensing, but also in data processing, functional safety, among many other factors.
What are some of the other hurdles we need to pass before fully self-driving vehicles are a reality for consumers?
Because there are so many sensors and many of them are running at higher resolutions and frame rates, there is so much data being used in autonomous vehicles.
Data processing is a challenge when prototypes are using central processor units that consume hundreds-of-watts of power. In-vehicle networks need higher bandwidths at the gigabit-per-second range. Then there is the issue of software that teaches the autonomous vehicles how to drive.
The “holy grail” in autonomous driving is to make such a complicated system more effective and reliable in performance, as well as making it efficient.
To make autonomous vehicles a commercial reality, you need to ensure that the self-driving system doesn’t use up so many miles of driving range (as they will certainly be electrically-driven).
If something does go wrong, a control centre operator could “step in” and drive the vehicle remotely.
You also need to ensure that users known how to hail a vehicle, how to make them comfortable and how these vehicles can communicate to them and other road users – so there will be considerable development needed in the user experience side of these vehicles.
How do you account for the price discrepancy between solutions?
I’m not sure how exactly Luminar is able to bring down the cost of its sensor over others, but it mentions the Iris sensor ‘platform’ it is leveraging to enable economies of scale with demand for its LiDARs from other industry sectors, in order to bring down unit cost for automotive. This sensor uses a mechanical method of scanning the Field of View.
However, LiDARs with a solid state design will bring about further cost reduction.
By minimising mechanical scanning methods, LiDAR developers can reduce size and dispense with electric motors and moving parts and, so, can also enhance reliability. Such solid state designs can include Flash LiDARs (e.g. ASC-Continental, Fastree3D, LeddarTech and Strobe-Cruise Automation), optical phased arrays (e.g. Quanergy and Xenomatix) and beam steering concepts, such as using Frequency Modulated Continuous Wave lasers (e.g. Blackmore-Aurora) transmitted through a smaller aperture. Many other LiDARs are still mechanically-scanning the Field of View, but with smaller micro-mirrors produced with MEMS (Micro-Electro-Mechanical-Systems) technology.
As more companies race to make competitive LiDAR systems, how can regulators effectively manage the future of driverless vehicles?
Regulators (or rather legislators) are concerned with consumers’ perception of autonomous vehicles and their lack of proven safety. Because of this, Strategy Analytics believes that autonomous vehicles are more likely to be first deployed in a “geofenced” environment, meaning they are only allowed to drive in a certain location on certain roads, under certain conditions and under strict supervision, before they are proven to be reliably safe and be allowed on public roads.
Photo by Alessio Lin on Unsplash.
Source of Article