Elon Musk vs LIDAR. Again.
Elon Musk enjoys a regular airing of his thoughts on electric vehicles, rockets, and boring machines. And LIDAR.
Here’s what he said just the other day at Tesla’s inaugural Autonomy Day event:
“Lidar is a fool’s errand. Anyone relying on lidar is doomed. Doomed! [They are] expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices. Like, one appendix is bad, well now you have a whole bunch of them, it’s ridiculous, you’ll see.”
We asked Michael Milford, Professor of Robotics at the Queensland University of Technology, about his thoughts on Musk’s LIDAR lambast.
’Elon is always good for a stir. He is basically the last public tech figure who hasn’t backtracked significantly on the timeline predictions of autonomous vehicles.’
Professor Milford sees LIDAR as a valid, current part of the CAV technology mix because of its importance in covering for the current shortcomings of cameras.
‘Humans drive primarily through their vision. Typically, current AV systems use vision via cameras, LIDAR, radar, etc., partially because “more is better”, and also because our current vision-only systems aren’t as good as the human eye and the human intelligence behind it. Consequently, the current CAV systems work very differently to humans.’
‘There is some debate about whether it’s viable to take these multi-sensor, highly-engineered solutions the last few percent in performance needed to make them viable. If the answer to this debate is no, then at least two scenarios could come about. One, the whole field retracts and no-one speaks about it ever again, or two, Elon gets the last laugh and everyone goes back to first principles to design a system that works more like a human relying on vision from the ground up.’
‘However, in the short-term it’s likely that Elon’s predictions aren’t going to look good.’
Before we go into what LIDAR does, and does not do, it’s also worth considering this. If one argument is the cost, yes, LIDAR is currently quite an expensive sensor. But that will not remain the case, Whether it’s due to volume increase, or to advances in LIDAR tech and manufacturing, the price is certain to fall.
“Many of these companies are competing on price for a problem that is not yet solved to a widespread commercially deployable reality. If the core problem isn’t solved sufficiently, these price differentiators will largely become moot.”
LIDAR’s current place in the CAV ecosystem
To further investigate the role of LIDAR in the mix, I asked Michael if the use of LIDAR is validated because it augments the capabilities of the CAV system’s visual components, much like the human brain adds and fine-tunes important information fragments for the overall acuity of the system? Or, as Elon Musk attests, the LIDAR component is simply an unreliable narrator.
‘LIDAR does indeed augment, particularly in doing things that machine vision isn’t great at. Such as judging the speed of an oncoming car on the highway when you’re waiting to merge, said Professor Milford.
‘However, while intuitively it’s a sensible thing to fuse information from multiple sensors, that intuition can let us down in some one-in-a-million situations. For example, on a curved highway, to not freak out about an incoming car detected by LIDAR that is on a trajectory to hit you, you might make the wrong decision. Visually, a human sees that the car is on the other side of a divider strip and makes some strong assumptions. The tech currently solves this problem through having prior maps and/or using vision to provide context to the LIDAR scan.’
‘Getting the balance between LIDAR and cameras just right in all situations is pathologically difficult.’
So the clock is on, Elon. Let’s see what the future of automated vehicle vision will be.
What do you think … is Elon right, wrong, or premature on this? Let us know your thoughts in the comments section below.
Discover more from iMOVE Australia Cooperative Research Centre | Transport R&D
Subscribe to get the latest posts sent to your email.
I think this is another view on the perennial question around detection technology and trying to find the silver bullet, which does not, and (in my view) will never exist. You need to remember, not all LIDAR is created equal, there are those that use an amplitude modulation (AM) approach to sensing, and a newer class of LIDAR that relies on frequency modulation (FM). The problem (argument) imitates from the market position that to date, over 95 percent of the $1.1 billion “LIDAR bubble” is invested in companies pursuing AM sensing. So it must be pretty rock solid, right? Not according to those in the FM market – what a surprise! and given that many major manufacturers (BMW, Audi, Toyota) and making the move to FM, you can see why the statement has been made. These companies are looking to the learnings of companies like Blackmore (https://blackmoreinc.com/) who have been do FM for the US Department of Defense for the last 10 years, so there might be something in this?
Depends on what kind of LIDAR technology is used. Baraja, an Australian company, fires lasers into a prism. As you may recall, light entering a prism will split into a rainbow. Each colour of light on the rainbow will come out at different angles to (a) the laser that went in and (b) the other colours of light in the rainbow. Fire the laser into the prism at sequentially different angles and the resultant rainbow that exits the prism will sweep from side to side. Do it at super-fast speeds and you have a very effect spectrum sweeper. Shift the wavelengths of light into the non-visible, make sure you collect the resultant bounce-back of light, tie it into a computer and you have a very efficient super-fast LIDAR scanner. And, this the key point, all the clever stuff takes place in a box inside a vehicle. The scanners themselves are made from cheap mass-produced glass made for the telecomms industry. I wrote an article about it for my employer FreightWaves. It’s a bit long but it explains everything you need to know about the company and the tech right here –
https://www.freightwaves.com/news/technology/machines-see-with-rainbows