digest | From the Field: 3 key obstacles for self-driving cars

Hidden limitations.
November 10, 2018


image | above
A town’s autonomous shuttle project.

photo | by PeachTree Corners


— the feature —

platform: Medium
blog: Scott Friedman
story: For self-driving cars, there are at least 3 giant hidden limitations
author: by Scott Friedman
date: October 2015


note: collected for library


Some first-hand insights.

In robotics, you have to look for the hidden limitations. They’re there, but robotics scientists and the press don’t talk about them, and civilians don’t even know to ask. My name is Scott Friedman and I founded the company Seegrid with pioneering engineer Hans Moravec PhD — plus with generous support from futurist + inventor Ray Kurzweil — who was on our Seegrid board of directors. So I’ve got a little expertise here.

For self-driving cars, there are at least 3 giant hidden limitations:


no. 1  |  lidar

All the current prototypes for self-driving vehicles are dependent on lidar sensors to know where they are, and to understand objects. The word lidar stands for: light + radar. Lidar shoots laser light at objects. When it bounces back, you know something is there. This works magically. Except when it bounces off of a raindrop, or a snowflake, or fog, or perhaps even smog. In these conditions, when there is something physical in the air, lidar is pretty much blind.

Other sensors can help muddle along — but that currently is far beyond the state-of-the-art. Probably the furthest along is Mobileye by Intel — but it doesn’t look like anyone is using it in the field yet, for fully autonomous driving  —  that is code for: it doesn’t work reliably. If it worked reliably people would use it.


no. 2  |  maps: real time + historic

This is a very complicated topic. You need both kinds. Only Google is doing any actual research on maps for self-driving cars. They are spending a lot of money — and have some of the top people in the world on the project. Everyone else is just wasting corporate research and development dollars for press coverage. Here’s some candid info on this:


on the web | reading

Technology Review • by MIT | Hidden obstacles for Google’s self-driving cars
deck: Impressive progress hides major limitations of Google’s quest for automated driving.

blog • on Medium | The robo car wars
blog • on Medium | Self-driving cars are way more complex than popular media would have you believe


no. 3  |  insurance

Who will get the law-suit when a child dies in an accident with a self-driving car? The answer is: everyone involved — especially the company with the most cash. So even if the automated car is parked: if it gets hit, and someone gets hurt or killed, then the automated car company could be legally taken to court.

There is no established insurance or liability law for this, it’s going to take a long time to solve. The only way through is for the government to give the self-driving car industry a full waiver of all liability. This is possible.

The former US president Richard Nixon did it for the Health Maintenance Organization — HMO — industry. Not only was its launch required to control health care costs, but this allowed the HMO industry to grow + thrive.


A final note.

None of these problems exist for flying cars. The technology that ships in $800 drones is just about good enough to launch the self-flying car industry. For flying cars, it’s about energy. So people, keep on working!


on the web | reading

Road Show • by cNet | Ford patent application cleverly hides lidar in side mirrors
deck: Maybe self-driving cars won’t be covered in awkward jumble after all.


by definition | What is lidar?

The word lidar is a portmanteau of light + radar. It’s also called: light imaging, light detection, and light ranging.

Lidar is a map-making technology — for geography + objects — that measures distance to a point by lighting-up the target with pulsed laser light — and then measuring the pulsed with a sensor, when they bounce back. Differences in the time it takes to for the laser pulse to return to you, can be used to make digital 3D models of what you’re trying to see.

Lidar also called 3D laser scanning, a special combination of a 3D scanning and laser scanning. It has ground, environment, and mobile applications. Lidar is good for making detailed maps: for geography, cartography, geology, forestry, weather + earthquake studies, and airborne industrial + military projects. Lidar is outfitted on self-driving vehicles to navigate roadways.


on the web | pages

Mobileye • by Intel | home
Mobileye •by Intel | YouTube channel

Seegrid | home
Seegrid | Hans Moravec PhD
Seegrid | YouTube channel


— notes —

MIT = Massachusetts Institute of Technology
HMO = health maintenance organization
cNET =

* Richard Nixon is Richard Milhous Nixon
* Hans Moravec is Hans Peter Moravec PhD
* Ray Kurzweil is Raymond Clyde Kurzweil

* MIT is org.
* Intel is co.
* Seegrid is co.