Safety First For Automated Driving

We were recently given a copy of “Safety First For Automated Driving 2019”. It is an attempt by a consortium of AV stakeholders to standardize and formalize the validation and verification of testing for AV systems. It was authored by representatives from Aptiv, Audi, Baidu, BMW, Continental, Fiat-Chrysler, HERE, Infineon, Intel, and Volkswagen, so a good cross-section of the different players in this industry. A copy can be found here.

 

 

 

 

There is a lot to this document. We won’t attempt to cover it in any detail. But there are three areas of particular interest to those of us in the work zone safety world:

  • The hand-off of vehicle control from the automated system to the operator.
  • Fail safe procedures when the system encounters something it does not expect or understand.
  • The importance of an accurate, real-time digital map.

VEHICLE-INTITATED HANDOVER

The moment the system recognizes that the conditions it “sees” are not the same as what were expected at that location, “the system shall react to compensate or shall issue a driver takeover request with a sufficient time frame for the takeover.” If the vehicle operator is unable to take control for any reason the system must initiate a fail-safe maneuver.

FAIL-SAFE PROCEDURES

“The automated driving system shall recognize system limits, especially those that do not allow the safe transition of control to the vehicle operator, and react to minimize the risk.” In the event something unexpected is encountered and when it is dangerous or impossible to hand control to the operator, the system has three options:

Comfort Stop:

This sounds similar to the vehicle pulling over when safe and parking out of the travelled way.

Safe Stop:

This would involve somewhat harder braking and the end of operation may be off the road or potentially could occur while still in a live lane.

Emergency Stop: Just like you locked up the brakes when traffic suddenly stops in front of you. There is no time to do anything else. This would leave the vehicle in the live lane. But we assume the condition causing it would be apparent to the other drivers, as well, so they would likely stop, too.

IMPORTANCE OF ACCURATE MAPS

The report states this clearly, “Failures relating to planned road changes can be avoided by incorporating road change plans from a road authority into the map updating process.” This, of course, requires real-time updates to the data. Like most people outside of our industry, they assume perfect knowledge by the DOT. But as you know, most “road change plans” are often wrong. Many contractors file provisional traffic control plans. Start and end dates change, weather delays the work, etc.

Now it was not within the scope of this document to address the how, just the why. It is up to us to find the most efficient ways to update maps in real-time. But they did recognize the need, and they also felt it was important that everyone have access to the same information. “as indispensable public information, road construction and maintenance plans should be fully transparent and easily accessible by all map providers.”

They also addressed the need to find out when the maps are wrong and to learn from those incidents.  “Errors as a result of real-world changes are difficult to monitor and control, thus they should be carefully analyzed.” Our industry will need access to the data so that we can optimize the ways in which our systems define and report road changes.

Finally, they reiterate the importance of accurate, real-time information, “an effective mechanism for map updating or maintenance is critically important” (emphasis added).  We are happy to see they recognize the need, but they don’t understand how to accomplish that: “A map updating or maintenance platform that comprises sensor data collected from multiple inputs, including but not limited to survey car fleets, massively deployed intelligent vehicles (e.g. vehicles with the ability to collect sensor data), high resolution satellite images and/or road infrastructures with surveillance sensors, can effectively detect the random road changes and lower the risk of random RMA failures.”

The good news is that they recognize the same AV safety issues we have. The bad news is that they do not understand work zone reporting. They believe agencies have perfect knowledge. Any gaps in that knowledge they still think can be filled by probe vehicles gathering and reporting changes as they are encountered. While that will help in some regards, work zones must be automatically reported by the people or equipment doing the work. No other method changes the maps accurately and the moment the lane is closed or reopened.

FHWA Seeks Pilot Sites for Work Zone Data Framework

 

In the Spring Edition of the FHWA Work Zone Management Program newsletter, they included a notice saying the FHWA Work Zone Data Initiative is looking for state agencies willing to use the Initiative’s  new work zone data framework to collect, process and share data.

This initiative is important for a number of reasons, but primarily because until we have a national standard saying what data should be collected and how it should be stored and shared, we will not be able to compare work zone safety and efficiency from state to state. Every state truly is different, but by opening these doors each state will be able to learn what they do especially well, and where they might improve.

Todd Peterson is the FHWA manager for this project and we encourage states to talk with him to learn more about this effort and how they might get involved. But let’s take this a step further.

If you are a state agency who is already working with Connected Work Zone systems please, PLEASE get involved in the initiative.  The Work Zone Data Initiative is working to identify what data should be collected. If you are already working with Connected Work Zones you have a wealth of data at your fingertips. And if you have been doing it very long, you already have a good idea of what is valuable and what is less so.

Most states don’t know what is available. They have their own databases and not much else. So, we need your participation to be sure those types of data are included. You also have some experience with the format or formats for that data, and you may be able to offer good advice on the best one for this application.

This effort has been many years in the making. It would be a shame if it is done without your help. We might even have to scrap this new framework and start over if we don’t get it right the first time. Call or write Todd Peterson today. And learn more about the Work Zone Data Initiative by clicking on the links.

Required Operator Training for Autonomous Vehicles?

We have talked here in the past about the difficulties autonomous vehicle drivers (operators?) will have acclimating when control of their vehicles is handed back to them, such as when they approach a work zone. Studies with simulators have shown a need for anywhere from 4 to 14 seconds for a driver to get a full grasp of all of the relevant external factors they must consider as they begin to drive.

A recent article in Axios Autonomous Vehicles points out that aviation has made use of automation for some time now. And they, too, understand the problem of moving from automated to human operators. In aviation, training focuses on that hand-off. Pilots are drilled in flight simulators on a variety of potential problems. So, when they encounter that problem during a real flight, muscle memory takes over and they react quickly and correctly.

The recent 737 Max 8 crash further underlines the importance of that training. It was apparently not included and that may have contributed to the pilot’s difficulty in regaining control.

The difference between aviation and autonomous vehicles is that training is mandatory for all pilots. If you fly a 767 you must stay current in all 767 training. However, for vehicles, a big selling point is that drivers no longer have to drive. They are told they can act more as passengers – gazing out the window, catching up on work, or watching an endless variety of streaming entertainment. Getting from that idea to one of mandatory training is a very long stretch!

Adding to the problem are the very different ways automakers are designing the machine-to-human hand-off. Each one is different.

In the Axios article, they quote Steve Casner of NASA, “We’re terrible at paying attention — and we think we’re awesome at it” Mr. Casner argues that drivers will need training. And they will need continuous updates to that training in order to learn how to deal with automation. Without initial user training and frequent refresh classes drivers will quickly become complacent.

This is a new topic of discussion but one that we must have to make CAVs safe for work zones and other segments of roadway with changing conditions.