Traffic Engineering

Many individuals and organisations (especially companies like Uber) claim that CAVs are the near future solution to all of our mobility and road safety problems. Leaving aside the issues around potential increased traffic congestion due to extra trip making, and the moral dilemmas about prioritising which persons are to be saved in a crash situation, how will CAVs deal with situations commonly found in typical inner city areas?  These situations include (but are not limited to):

1.     Narrow streets where only one vehicle is able to proceed in either direction at one time
2.     Badly faded signs
3.     Obscured signs
4.     Damaged signs
5.     Non-standard sign
6.     Badly faded or non-existent road marking

The following examples are taken from an approximate one square kilometre area in Camperdown, in Sydney’s inner west. This is not to suggest that the local council is necessarily negligent (or much different to other LGAs) in their maintenance routines – rather that these situations are mostly correctly interpreted by experienced drivers, and therefore may not be safety critical. CAVs, however, must be programmed to make ‘black or white’ decisions – there is no room for interpretation.

Narrow streets where only one vehicle is able to proceed in either direction at one time

This two-way street has a 50 km/h limit, with parking on both sides.  Experienced (and especially local) drivers look well ahead to determine if there is any oncoming traffic, and then make decisions about who will go first, and who will seek refuge in a kerbside space or side street entrance. Good drivers make eye contact and thank each other for their cooperation. What would CAVs do?

Badly faded signs

These are examples of badly faded NO PARKING signs in rear lanes.  While it might be argued that CAVs are unlikely to be using such lanes, there may be some instances.  Would the CAV know not to park here?
 

Obscured signs

Believe it or not, there is a STOP sign at the intersection with the cross street, but it is completely obscured by the foliage (see next photo). Note also the lack of stop line and holding line at the intersection. How would a CAV interpret this situation?

In this photo, can you see the STOP sign?  The next (zoomed) photo shows just the edge of the sign protruding past the building line. Again, the stop line and holding line is badly faded.  Would the CAV know how to proceed?

Non-standard signs

There is an unusual combination of information on this pole. The non-standard sign is “Trucks use Derby Place”, due to the tight corner at the STOP sign. How would the CAV interpret this situation?

Badly faded or non-existent road marking

These badly worn road markings are intended for cyclists, indicating that the cycle route goes around the corner to the right. The arrow symbol is less worn than the bicycle symbol. Might the CAV interpret the arrow as a compulsory right turn?

Challenging non-standard signs

A RESPONSE TO ALAN FINLAY'S DISCUSSION ON 'HOW WELL CONNECTED AUTONOMOUS VEHICLES (CAVs) DEAL WITH THESE SITUATIONS?'

Graeme Pattison

In contrast to Alan's distinct unease about how future autonomous vehicles will cope with tight inner city local roads, poor signposting and poor road markings, I believe CAVs will perform more than adequately. Alan also mentions a moral dilemma about prioritising which persons are to be saved in a crash situation. I see this CAV issue not as a dilemma but as an improvement on the present situation. The explanation is that in a crash few drivers have anticipated it more than a couple of seconds in advance and then find they must intensely focus their attention on controlling the vehicle to limit crash severity. If there is any remaining spare attention capacity the driver probably only has one second or so to determine which people are at risk of injury, who they are and what their perceived priorities (eg child vs adult, innocent bystander vs crash contributor) and then predict the possible trajectories of all involved. A very complex problem to solve in a second if you are human but not as  challenging if you are a high speed automated processing system with more data on the situation than any human would have. In addition, the CAV will make the "who to save" decision based on algorithms that have been extensively planned, perhaps from hundreds if not thousands of human hours of discussion, thought and evaluation. 

Yes inner city local streets are often not ideal, may not be up to standards and generally have some signs that may be faded, obscured, damaged, non-standard or even missing. Road markings can be badly faded or non-existent too. But CAVs and the road system will be designed to overcome most of these problems. Alan states that CAVs "must be programmed to make 'black or white' decisions - there is no room for interpretation". My view is that to operate in the real world CAVs (which will never have perfect information on the whole road situation) will cope by applying risk management and the Precautionary Principle. This means they will still proceed in unclear (or yet to be determined) situations but at a speed or on a path that sufficiently minimises any risk of an undesirable incident. We all do this in our lives and should accept it as a valid process for an autonomous machine.

Alan also correctly states that experienced drivers can mostly correctly interpret signs and marking that are faded, damaged or otherwise in very poor condition. But inexperienced drivers, the non-locals and those who are distracted or not paying attention may do badly. So poor signs and markings are a hazard for many drivers, at a level that may be higher or lower than for CAVs. Some autonomous vehicle proponents and road agencies are proposing that all road and traffic facility infrastructure be fully digitised with the data sets available on-line for real time CAV use via 5G communications. It is expected that this will even be a legal requirement in some circumstances. An example is a temporary gravel detour alongside a highway undergoing road works. The detour would not be shown on normal GPS maps, the edges would be difficult to visually determine and signposting may be poor so CAVs need the extra digital site information. The data preparation would likely be a mandatory part of the road works approval process. In developed countries almost all road infrastructure and traffic devices would already be recorded in digital databases and CAD files. There is of course work to be done to transfer this information into compatible and accessible formats and then keep it up to date.

Alan's first photo shows a narrow two way street with only a single driveable lane width between parked cars. Driver cooperation is needed for vehicles to travel in opposing directions. But autonomous systems can also achieve this through their programming. I have seen it happen with Honda's Asimo robots moving towards each other in a game of soccer at UNSW. See robot soccer at https://youtu.be/4uYN_3gL4_Y. Judging by company approaches, an Uber CAV is likely to be more assertive than an Apple CAV and make its way through first. Two opposing CAV vehicles may also communicate with each other and decide which proceeds and which gives way based on a built in cooperation algorithm. At a recent AITPM presentation Tim Armitage from the UK described his on road experiences in a self driving vehicle. He found it to be more risk averse and conservative than his own driving manner.

The faded and obscured signs shown in other photos give the same problem to humans and CAVs. Many humans will ignore illegible signs and markings and proceed safely, with caution if needed. CAVs will do the same but if there is a trustworthy complete on-line signs and marking database then CAVs have the potential to know what should or was there and abide by it, making their performance superior to that of human drivers.

Non-standard signs present a different problem as they may be legally mandatory or they may give ill-advised directives. Again, I see CAVs proceeding slowly using the Precautionary Principle. But in any case, it is unreasonable to expect CAVs to fully resolve ambiguous information that a human is not able to decipher. The CAV will probably take a more rational calculated approach than a human. In any case CAVs will be a different animal to humans and will require some system changes such as the removal of ambiguous and illogical signs.

In summary I see advantages with CAVs taking an analytical approach, never being tired, emotional or impaired as humans may be. CAVs will never be perfect and fatal errors will occur but I expect the rate to be low. Tesla's Elon Musk claimed in a May 2018 tweet that Tesla's cars have a fatality rate four times better than the USA's one every 86 million miles of driving.

Perhaps we will need ANCAP style testing to rate the various vehicle manufacturers' software and control systems.