I wish there’d be more initiatives to make roads “legible” by self-driving cars: distinctive computer-readable markings, machine-readable codes on signs, etc.
There's some interest in that in China. The US has assumed that self-driving cars get no support from the highway. With some minimal support, it's easier. Volvo wanted to drive magnetized nails into roads to mark them under snow.
this is an interesting and prescient comment for a couple reasons...
1) road sign standardization does exist, it is called the MUTCD [0] and SHOULD be highly usable for this type of training of image systems. I would be shocked (really shocked) if there are people working on this who have not loaded the full size EPS graphics into their software in some way [1]. However, when you install billions (I'm WAGing there are probably billions of road signs in the US?) of signs - variance is problematic. The signs themselves can age, wear, or be wrongly printed. They can also be effected by situational variables such as blocked by a tree limb, installed incorrectly (misaligned), or be temporary, or be affected by weather (snow, ice, rain). Standardizing is very much done but has its limits.
2) Your examples of 'legibility' is really interesting because it presumes human sense - e.g., primarily relying on sight. There are lots of ways to make roads signs legible that rely on senses humans do not possess but computers can. Rather than putting code on signs that relies on vision, it would make much more sense to satisfice other sensing methods for communicating/broadcasting information about roadways. You could put long distance RFID in the signs, hell this seems like an actually good use of IoT. It would help solve the visibility, weather, aging, and temporary signage problems. Prior to the current age of self driving cars, this type of non-vision, non-self contained approach, roads designed for self driving cars was considered the most viable [2, see pages 18-19]. Good maps loaded into your onboard systems make a road more legible - even if they can't be seen. There are already ways to hack non-visible ways of making the road system 'legible' [3]
3) The last piece is that there is a huge difference between making the road signs themselves legible and making roads legible. [2, section 5.1, pg. 107] talks about this extensively and our perception of how we think about sensing as people and how we ontologically categorize vehicles as independent of the road...Its why you some people have laughed at the idea of working to better prepare roads for self driving cars [4], even in ways that were used throughout later 1900's attempts at self driving cars. for purposes of the current US road network that is probably a good assumption. Somewhere like Japan, you can probably assume the roads are much more consistently designed, signed, and implemented. Unsurprisingly, the first actual for sale level 3 car is in Japan [5].
[2] https://cmsw.mit.edu/wp/wp-content/uploads/2016/06/268475427...
footnote 28: In her book Magnetic Appeal, Joyce argues that “seeing does not equal truth or unmediated access to the human body,” but that practices equating these are so common that images are often used to stand for truth despite doctors’ awareness of how social practices shape this evidence [Joyce, 2008,p. 76]. Popular narratives are particularly prone to fall prey to the “myth of photographic truth” [Joyce, 2008, p. 75]. These tendencies are of great relevance when considering other complex, technological projects dependent on imaging and which use images rhetorically, to stand for the “truth” of their ability to perform a task—such as detect a pedestrian in a crosswalk.
Forget signs. An auto-drive car needs to respond to hand signals from people like crossing guards and road crews.
Ultimate test: a self-driving capable of loading itself on/off a car ferry. It will have to actively disregard signages/road lines to obey hand signals and announcements from the ship's crew. Will a tesla park itself within 6" the car ahead?
yup! signs are one component of information that a self driving car needs to operate in an uncontrolled space.
Even for a human, city/uncontrolled driving is hard mode.
A self-driving car needs to read human body language...people do it all the time while driving. Is that driver looking at their phone? is that homeless person going to do something aggressive or are they just pan handling?
I sort of intend a more general sense of “legible” here: something like the sense used in Seeing like a State.
But, in general, I think the sorts of in-road things you’re suggesting are really valuable: “upgrading” our roads this way should pay dividends in terms of the amount of capital it takes to get a new self-driving-car system off the ground: embedding such tech into the roads seems like a public good that would benefit anyone trying to make a public car by lowering the R&D cost and centralizing some of the intelligence necessary for such a system.
ha! one of my favorite books...just made a slide for a workshop to help engineering academics understand the difference between legible and legitimate.
Travel route 66. They had such trouble with people pulling down those signs that they instead painted them onto the road. But that doesn't work so well in bad weather as rain/snow makes such signs difficult to read. Painted road is also more slippery.
I wonder if you could embed little iron strips in the road in a specific pattern and then put some sort of sensor inside the tires to read the pattern. And then have a warning system to alert people about the transition from “upgraded” to “normal” roads.
