Monday, October 3, 2016

2016 Update to SAE J3016 Taxonomy and Definitions for Terms Related to Automated Driving Systems

SAE International has provided an update to SAE J3016, "Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles." The original version of the taxonomy, better known as the "levels of automation," was published in 2014. The revision was published September 2016.

This document has taken on increased importance now that U.S. DOT NHTSA has adopted the standard into federal policy

The scope of the taxonomy applies to “driving automation systems.” This is a change in language from previous version, which used the term automated driving systems. Automated driving systems (ADS) are still used in the taxonomy but now include only levels 3-5.

Driving automation systems are defined such that they “perform all or part of the dynamic driving task (DDT) on a sustained basis. The inclusion of “sustained basis” is new to this version and means that intervention systems such as ABS, ESC, and automated emergency braking (AEB), are not covered by this taxonomy. The standard states that “due to the momentary nature of … active safety systems, their intervention does not change or eliminate the role of the driver in performing all or part of the DDT, and thus are not considered to be driving automation.” A vehicle with AEB but no system that performs sustained control of lateral or longitudinal motion would be considered to have level 0 automation in this taxonomy.

The taxonomy (levels) apply to the driving automation features—not the vehicle. So a given vehicle may be equipped with a driving automation system capable of providing multiple driving automation features that perform at different levels. For example, a vehicle may have a driving automation system capable of levels 1, 2, and 3 driving automation, but it would not be correct to refer to this as a “level 3 vehicle.” The correct use of the taxonomy would be to say that the vehicle has a level 3 driving automation system.

The levels of driving automation are defined by reference to the designated roles of the driver and driving automation system. (The word, "driver" in this standard always refers to a human user.) In other words, even if a level 2 driving automation system is capable of monitoring the roadway while engaged, it is not considered a level 3 system unless the manufacturer has designated it as such.

Note: The summary provided here is only of the document scope, which is in the public domain. The entire document has been made free to SAE members, but I'm not sure if I'm allowed to re-publish parts of the document body (such as definitions) without permission. If I can, I'll provide another update. Meanwhile, SAE members can download the entire standard here: 

Wednesday, February 10, 2016

No, NHTSA did not just legalize robotic drivers

In November of 2015, Chris Urmson, Director of Google’s Self-driving Car Project, sent a letter to the National Traffic and Highway Safety Agency (NHTSA) of the USDOT requesting clarifications of how Federal Motor Vehicle Safety Standards (FMVSS) would apply to a self-driving car (SDV) designed to be operated only by an automated driving system (ADS).

In other words, Google proposed building a vehicle with no manual controls at all, and wanted to know if NHTSA would allow it.

NHTSA's response clarified that the FMVSS does not explicitly require a human driver. For NHTSA’a purposes, FMVSS requirements that reference a driver or operator could be faithfully interpreted as referring to the SDV itself. But NHTSA was willing to go only so far in that interpretation.

In response to a list of FMVSS provisions that Google submitted requesting that the SDV be considered a driver, NHTSA highlighted some specific rules that could not be met. The rules that impose problems are generally those that imply the driver is a human person. For example, FMVSS No. 135 requires brakes to be activated by a foot control. Thus, a vehicle without a brake pedal could not meet FMVSS No. 135. There are other provisions that govern vehicle features in relation to a driver’s hands and eyes.

A fun interpretation of NHTSA’s argument is that Google could completely meet FMVSS by inserting a human-sized android into the left-front seat of a traditionally-operated vehicle and calling that robot the driver. But NHTSA acknowledges that if the robot driver is the SDV itself, many of the FMVSSs just don’t make sense. NHTSA was unwilling to adopt interpretations that clearly contradict the language of the FMVSS (such as the requirement for a brake pedal), but the agency also implied it does not intend for FMVSS to be a barrier to SDV deployment. NHTSA suggested that it would be investigating future rulemakings that would allow SDVs like Google to comply with Federal regulations. In the meantime, NHTSA suggested that Google go ahead and develop an SDV without manual controls, and petition to have it exempted from certain provisions of the FMVSS.

Within Federal regulations, automakers can request for temporary (up to three year) exemptions from safety standards. The exemptions are frequently requested, and usually granted, for very minor reasons of non-compliance. For example, a status light that fails to activate in unusual situations, or a low-volume carmaker receiving a one-year extension before it has to meet a new crash standard.

Requests for exemptions to entire standards that do not apply to a robot driver would be orders of magnitude more substantive than the usual minor exceptions. NHTSA has invited Google to explore this option, but has not indicated a likelihood of success. Receiving permission to sell a vehicle without manual controls in the U.S. is by no means assured, and NHTSA has made that clear. In this letter of interpretation to Google regarding self-driving vehicles, NHTSA has pointed to a path to deployment, but by no means paved the way.

NHTSA says computer can drive car for FMVSS tests, maybe

The big news today in the world of self-driving vehicles (SDVs) is that NHTSA has published a letter sent to Google in response to clarifications regarding federal safety requirements. The popular press is interpreting this as “US tells Google computers can qualify as drivers.” Technically, this is true, but also misleading.

First, if you are a US citizen with a driver’s license, take a glance at it. I haven’t confirmed this for every state, but I almost guarantee that your license does not reference federal law or NHTSA in any way. USDOT does not license drivers. States do. NHTSA’s opinion on who/what qualifies as a driver does not clarify responsibilities or liabilities of manufacturers and/or human drivers/occupants of SDVs in any way.

NHTSA’s authority in this area is limited to promulgating and enforcing the Federal motor Vehicle Safety Standards (FMVSSs). These rules describe safety requirements that a vehicle must meet to be legally sold to consumers as a registerable vehicle in the US.

As you might expect, several of the FMVSSs describe how vehicle controls must function in reference to a driver. However, the driver is always assumed to be a human person. This assumption is implicit. The FMVSSs to not state that explicitly that the vehicle must be driven by a human. For example, FMVSS No. 135 states that vehicles must have a braking system activated by a foot control. Google requested that NHTSA consider how it would interpret that requirement if the driver were non-human (i.e., an AI).

NHTSA agreed with Google that the driver could potentially be non-human. NHTSA also declared that FMVSS No. 135 was clear enough that to meet the requirement, there would have to be a foot pedal that activates the brakes. However, NHTSA acknowledged that FMVSSs were not written with AI drivers in mind, and implied that the agency has no interest in forbidding self-driving cars as a matter of policy. As such, NHTSA said that the agency would review and FMVSS standards could be revised to allow SDVs to meet the standards.

More importantly, NHTSA directed Google to a provision of federal law (49 U.S.C. 30114 and 49 CFR Part 555) that allows automakers to petition for exemptions to FMVSS if they can “demonstrate that features of their products provide equivalent levels of safety to those provided by FMVSS.” NHTSA essentially encouraged Google to go ahead and build an SDV, and pledged not to obstruct deployment if everything seems relatively safe. But NHTSA has not guaranteed that a petition for exemption will be successful.

Another point that should be mentioned, is that NHTSA reiterated (as they have in the past) that the agency “does not make determinations as to whether a product conforms to the FMVSS outside of a NHTSA compliance test.” In other words, if it passes NHTSA test, it can have any level of self-driving technology layered on top of that.

So, Google’s proposed SDV poses a potential problem, because many FMVSSs and NHTSA compliance tests directly reference the performance of the vehicle in relation to a human driver. Unless Google successfully petitions for exemptions, it appears that, yes, a FMVSS-certified vehicle will need a footbrake. (Interestingly, a steering wheel and accelerator don’t seem to be a requirement.) However, if a car is able to meet normal requirements as a manually-driven car, there is no hard-stop that would prevent that vehicle from having an autonomous mode in which all aspects of the driving task are assumed by the vehicle itself. Cars could even be sold that could be deployed without human occupants.

But to go back to the beginning of this post, NHTSA has authority over what is sold, but US states have authority over how vehicles are used. States could prohibit occupant-less vehicles, or even self-driving mode entirely. As it is, the four states that have addressed SDVs in regulations have explicitly made the human person who activates automated mode the driver and operator for legal purposes. Of course, since we haven’t really seen a deployment of SDVs in the real world, we still have no idea how regulatory agencies, police departments, and the court system will interpret the existing rules or promulgate new laws and regulations. NHTSA’s letter to Google was interesting, but didn’t really change anything.

Yeah, OK then.

Friday, October 17, 2014

Insomnia Book Review: Candide by Voltaire

I am an insomniac. While this sucks, it gives me a chance to read things that I wouldn't normally read. Usually, I’m hoping the text just puts me to sleep. But sometimes, it doesn't. This is how I came to read Candide by Voltaire.

I am not familiar with Voltaire’s work. I know he was writing in the eighteenth century and associated with Enlightenment philosophy. I assumed Candide would be a work of philosophy. It isn't. It’s a short novel written by somebody who was clearly fatigued with elaborate philosophical arguments. It’s a drunken rant.

Candide is the name of a man. Candide was born into minor nobility and enjoyed a fine childhood. He had the comforts of castle-life, including the tutelage of Pangloss, a professor of “metaphysico-theologico-cosmolo-nigology.” Pangloss explained to Candide that God is perfect. Thus, the world is perfect, and is as it must be. In fact, this is “the best of all possible worlds.”

The philosophy of Pangloss reflects beliefs in an omnipotent and wholly benevolent god that existed in the eighteenth century and still do today. Pangloss’ philosophy is rekindled every time one utters the phrase, ‘God has a plan,’ or, ‘Everything happens for a reason.’

Modern philosophers have pretty much concluded that nothing happens for a reason. There is no metaphysical logic to the universe. If God has a plan, it is often a really shitty plan.

Back in Voltaire’s day, there actually were learned and reputable men who argued extensively that humans live in the most perfect of all possible worlds under a all-knowing, all powerful, and perfect God. After the Reformation, it became possible to question religion without immediately being executed. A few brave souls experimented with asking questions such as, ‘if God is powerful and good, why does evil exist in the world?’ Those, like Pangloss, who subscribed to this belief, reasoned that there must be evil in the world to allow humans to exercise free will. It’s a little more complicated than that, but it really doesn't matter. Anyway, plenty of people still hold such beliefs.

Voltaire clearly disagreed.

Voltaire’s Candide was banished from the castle after innocently feeling-up the 17-year-old princess. He was forcibly conscribed into the Bulgarian army. Not wanting to be in the army, Candide expressed a wish to exercise his God-given free will to not go to war. The army did not consider this a valid use of free will. Candide was tortured and beaten within an inch of his life. An official execution was arranged to take the final inch. Moments before Candide was to be shot,
“The King of the Bulgarians passed at this moment and ascertained the nature of the crime. As he had great talent, he understood from all that he learnt of Candide that he was a young metaphysician, extremely ignorant of the things of this world, and he accorded him his pardon with a clemency which will bring him praise in all the journals, and throughout all ages.”
Candide was patched up and put back on his feet to be marched into battle. In battle, Candide “trembled like a philosopher [and] hid himself as well as he could during this butchery.” He slinked away after both armies had effectively killed each other off.

By chance (or design), Candide met his former professor of metaphysico-theologico-cosmolo-nigology. Pangloss was ridden with syphilis and begging on the street. He reported that the castle and been sacked in the war, and the royal family murdered  (including Candide’s beloved princess). In due time they experienced a great earthquake that convinced them that end-times had come. It did not. Though death was all about, it was not the coming of the Kingdom of God, it was simply life on Earth. For all these misfortunes of disease and war and natural disaster, Pangloss retained his optimism, saying, “for private misfortunes make the general good, so that the more private misfortunes there are the greater is the general good.” God has a plan.

To make a short story shorter, the cast of characters are subsequently made to withstand and witness horrors unimaginable. Voltaire describes the most vulgar and wretched acts (most of which are historically accurate) with clinical detachment. People who have lost wealth and standing are portrayed as particularly pitiful. Voltaire seems to believe that a loss of riches and social status is among the most depressing situations in life.

The story ends happily, with Candide living peacefully with Pangloss and a small group of friends on a subsistence farm. The characters come to realize that there is no grand design. God is not testing us. Life is random and often terrible. Life is made more terrible by greed and deceit. The path to happiness is to be humble and kind. Though Voltaire has no respect for God, this is a very Christian message.

Wednesday, August 13, 2014

Why did Detroit Commuters get Caught in a Flood? (It wasn't because Rain.)

The initial response to the rainstorm in Southeast Michigan on Monday 8/11/2014 is that it was a freak occurrence—unpredictable and unavoidable. In some ways this is true.

When you get a lot of rain, things flood. In floodplains, this is unavoidable. This is why FEMA publishes floodplain maps that show areas predicted to be underwater  ever hundred years or so. But Detroit is not in a floodplain. Flooding problems in Detroit were not caused by a storm surge or rising rivers. If there was no city here, there would not have been a flood. The reason areas in Detroit flood is because we either dig basements which very easily fill with water, or build streets that don’t let water drain. As in most older cities, Detroit’s stormwater infrastructure is a disaster and it’s completely expected that a lot of rain will flood the surface streets. So, yeah, OK then, that will happen.

But let’s focus on the Interstates.

Interstates are designed to very demanding standards. The federal government requires high standards on Interstates because it is acknowledged that they are strategically and economically essential to operate under all foreseeable conditions. The rainstorm on Monday August 11, 2014 was not a freak occurrence. It appears to be within the limits of a 100-year design storm that the Interstate should be built to withstand. Unless I’m misunderstanding something, it looks like MDOT should be designing for 3-inches of rain in a 2-hour period, and 5.5-inches of rain in a 24-hour period. From what I can tell from spot-checking weather station data, the 8/11 storm did not exceed that.

The fact that drivers were forced to abandon their cars on flooding Metro Detroit Freeways means that something went wrong somewhere.

I-696/I-75 Interchange

If I am wrong and Monday’s event did exceed the design storm, MDOT and MSP should have been able to predict flooding and closed affected highways before they flooded, or were at least prepared for the possibility on short notice (i.e., had crews ready to react if flooding was observed). It may seem over-demanding of MDOT and MSP to have this capability, but they invested in a state of the art Traffic Operations Center that is supposed to do just these kinds of things (e.g., monitoring traffic and weather conditions to anticipate and deploy mitigating strategies). They may never had considered that Interstates would flood, since it hasn’t been a problem in the past; but it is actually their job.

MDOT/MSP Southeast Michigan Traffic Operations Center (SEMTOC)

But again, it looks like the storm was within design parameters, meaning the freeways never should have flooded in the first place. One possibility is that Metro Detroit’s Interstates were not designed to withstand a 100-year storm. This does not seem likely.

The more likely possibility is that the Interstates did not function as designed. These things are usually built pretty well. But sometimes they fall into disrepair. Concrete crumbles, drain basins get silted in, etc. There are prescribed maintenance practices, but maintenance is usually underfunded. In the case of Detroit’s Interstates, many of which are built in below grade (in big troughs) the drainage system is heavily reliant on pump stations. My guess is that for some reason, pump stations were not in working order.

Whatever the cause, the fact that motorists had to abandon their cars to rising waters on Monday means that a mistake was made somewhere. This is a good thing, because we will get more rainstorms like this. The fact that it was a mistake means that we can find it and fix it. That process needs to begin as soon as the mud gets scraped off the roads.

Thanks, @Jake_Dudek

For crazy pictures, search Twitter for #detroitflood. Yeah, OK then.