8 SMIDSY – looked but not perceived; workload

SUMMARY – when driving the brain needs to process sensory information… this is known as the ‘cognitive workload’… as driving tasks increase in difficulty, the workload starts to increase… even relatively simple tasks create an incoming information stream that exceeds the brain’s ability to process it all… stress in driving tasks further reduces the brain’s ability to process data… once the workload limit exceeded, a driver’s situational awareness at junctions is significantly degraded…


Back in the early 2000s, on my ‘blog before they were called blogs’, I wrote about ‘workload’ and military helicopters and how much the pilot actually had to do whilst flying the machine, navigating AND using the weaponry. In trials, it turned out that even hugely increased automation wasn’t enough for one design to be a single-seater as planned. There was simply too much going on even for a highly trained pilot. As a direct result, the Comanche helicopter was modified to carry two crew.

One of the key theories behind research into car – motorcycle collisions is termed ‘gap acceptance’, which seeks to understand how the driver calculates the ‘time to collision’ with an approaching motorcycle, decides whether or not a gap ahead of an approaching motorcycle represents a safe distance, and then makes a decision whether or not to pull out. However, the research usually pre-supposes a straightforward task – that there are no distractions and having seen the motorcycle, the driver simply estimates the motorcycle’s distance and speed to calculate ‘time to arrival’.

It wasn’t hard to predict that workload would also be an issue for the typical driver (or motorcyclist, come to that), who doesn’t have a fraction of the training of a helicopter pilot. In the article I predicted that in a complex driving environment, the driver would experience high workload. A common solution, adopted by many animals including primates (the group that includes humans), is to process just a small area or a few objects at any one time. We can then scan the visual scene in small chunks, subjecting each to more detailed visual analysis. This scanning technique has been compared to a ‘virtual spotlight’, highlighting different regions and objects for a closer look.

But it all relies on being able to slow down the scan. Pammer et al (2017) noted in their conclusions that:

“When we are driving, there is a huge amount of sensory information that our brain must deal with. We can’t attend to everything, because this would consume enormous cognitive resources and take too much time.”

In other words, the complexity of the driving task could lead to a disconnect between eyes and brain and workload offers at least a partial explanation for ‘looked but failed to see’ collisions between a motorcycle and a car. Some of the visual information within the scene would simply not reach the conscious, thinking part of the driver’s brain. Focused on one visually-intensive task – perhaps searching for road markings indicating the correct lane on a busy, complex roundabout – other visual input goes missing and the driver loses track of the motorcyclist. Even when the driver looks in the direction of the oncoming motorcycle – in some cases appearing to look directly at the motorcycle – the motorcycle goes missing and the driver pulls out into its path.

driver-1149997_960_720

Picture yourself emerging from a side turning then turning right onto a busy road – the situation in which the classic SMIDSY collision occurs. We’re searching for, then monitoring, multiple moving objects, which are travelling at different speeds in at least two different directions – three if the junction is a cross roads. We’re looking in several different locations – the two lanes of the road itself, the nearside margin of each lane (where bicycles might be expected), plus pavements to check for pedestrians. We need to detect stationary objects which are in our lines-of-sight and be aware how they might create blind spots. We need to move our eyes from one search zone to another to visually acquire the targets to be scanned. Each eye movement (a ‘saccade’) takes time, and then the eye has to refocus on the new scene, which also takes time. And we should be keeping an eye on the mirrors too.  ‘Looking properly’ is far from a trivial task.

Stress is known to affect our ability to process information. Trying to follow complex road layouts in an unfamiliar town, we rapidly move towards a condition of stress where we are not able to process as much information. Once the workload exceeds the level we can handle, the results include what are termed “compensating behaviours”:

  • errors – we make the wrong decision
  • slow task performance – it takes a long time to reach the right decision
  • task shedding – we never make a decision
  • rapid task switching – we keep mentally jumping from one part of the overall task to another
Gap acceptance.png

In the worst case, we might fail to perform a task altogether, a condition known as ‘task shedding’. From working in motorcycle training, I know how novice riders who are experienced drivers can actually forget a ‘simple’ task like looking for conflicting traffic when pulling out of a junction.

Few of us are consciously aware of just how much workload even relatively simple tasks create. A study by Murphy and Greene in 2016 put forty two drivers into a life-size Volkswagen Polo driving simulator where they performed a series of gap perception tasks involving judging if their vehicle could fit between two parked vehicles. There were cars parked on either side of the simulated road. When the gap between the parked cars was easy to negotiate, 22 of 41 drivers noticed an unexpected pedestrian in a red blouse. But when drivers the gap was reduced so that it was barely wide enough for the car to negotiate, only seven noticed the pedestrian.

intersection-2262239_960_720

Passing between two parked vehicles is a trivial task compared with monitoring busy roads to decide if it is safe to turn. It should be clear that excess workload can significantly affect a driver’s situational awareness at junctions and in some instances a failure to spot other vehicles (including motorcycles) isn’t ‘lack of attention’, it’s ‘not enough to go round’. However, hardly any quantitative work has been done to investigate this. The authors point out:

“This study is the first to demonstrate perceptual load effects on awareness in an applied setting and has important implications for road safety”.

In September 2019, the authors of a paper entitled: “The ‘Saw but Forgot’ error: A role for short-term memory failures in understanding junction crashes?” proposed an important new mechanism for visual perception failure, suggesting that some motorcycle collisions which are current classified as ‘looked but failed to see’ crashes are actually the result of a memory issue – the motorcycle is actually seen and brought to conscious attention – but is then somehow forgotten again.

Robins et al (2019) began with the conventional explanation, the ‘looked but failed to see’ phenomenon:

“Typical interpretations of these junction crashes are based on the idea that the driver pulling out of the junction has failed to devote sufficient attention to the traffic on the road he or she is entering, thus, they are often termed ‘Look but Fail to See’ (LBFTS). It is proposed that the crash is caused by failing to spot an oncoming vehicle. This is consistent with the psychological phenomena of change blindness and inattentional blindness, with explanations suggesting that even when attention is on an object it is not always associated with the detection and processing of this object.”

I’ve always been very doubtful that drivers fail “to devote sufficient attention to the traffic” simply because fatal collisions are actually rare events – there’s more on this in the section ‘Looked, saw and turned anyway’. As the paper states “corresponds to approximately 90 deaths in the UK per annum”. If that sounds shocking, put it in context with around 1.2 active motorcyclists, travelling approximate three billion miles each year and meeting around 40 million drivers at a number of junctions no-one has ever counted. In that context, if’s obvious almost every driver must pay sufficient attention on nearly every occasion AND SEE ALMOST EVERY MOTORCYCLE EVERY TIME ONE IS VISIBLE. If that weren’t the case, we bikers wouldn’t get much further than the end of our own road.

They then mention what I have called the ‘looked, saw, and misjudged’ error:

“Other previous research has suggested that motorcycle accident risk is inflated due to the size-arrival effect, which suggests that smaller objects are perceived as further away, and to arrive later than larger objects. Due to this perceptual error, drivers may adopt a smaller gap at the junction when a motorcycle is approaching compared to a larger vehicle such as a car or large goods vehicle. Data from both real and experimental simulations have found that crashes do occur when a car pulls into the path of an oncoming motorcycle, with the car driver thinking the motorcycle is further away than it actually is.

As I’ve shown in previous pages, there is a third cause of junction collisions which wasn’t mentioned – the ‘looked but could not see’ issue where the motorcycle simply isn’t where the driver can see it, however hard he or she looks. As we saw previously, this issue appears to account for around one-fifth of all junction collisions.

The authors commenced a series of studies. And here’s something very important.

“Our study immerses the driver in a realistic driving scenario.”

Many of the earlier studies into driver perception of motorcycles were based on still photos, short video clips, or contrived on-road trials. Advances in computer generated images (CGI) now allows for a realistic driving task to be created for the subject but at the same time allowing the environment to be manipulated by the researchers. In this case, the studies involved created scenarios to examine whether drivers were willing to complete a manoeuvre in front of approaching vehicles – what is known as ‘gap acceptance’ – whilst using eye tracking to determine whether drivers had visually fixated on vehicles before subsequently failing to recall them.

So what did the paper find? They report that:

“The most striking finding from the first study was…the complete failure to report some vehicles, particularly approaching motorcycles.”

The authors proposed that we use short-term visual memory for “the encoding, temporary storage and retrieval of information for complex cognitive tasks”, including building an awareness of the traffic situation around us. What they suggest is that a driver’s ability to create a correct awareness of what’s around them “depends not only on [the information] being successfully encoded, but also on the storage of this information and its retrieval from short-term memory”, and that the so-called ‘looked but failed to see’ incidents “…might not always be due to a failure in visual attention (encoding), which many previous researchers have suggested, but could sometimes be due to subsequent failures in memory.”

In essence, they propose a fourth mechanism, which we might call ‘Looked, Saw and Forgot’.

Here is the interesting part:

“…failures to report a motorcycle were not predicted by how long a driver fixated on the vehicle, but were associated with their subsequent behaviour.”

The implication of that is that the current recommendations to look HARDER and LONGER for motorcycles is likely to be ineffective.

The authors also noted:

“One of the biggest challenges in this research was that such memory errors were rare – the majority of participants never made any memory errors at all.”

Once again, that’s absolutely in line with my contention that drivers DO pay sufficient attention to see approaching motorcycles on nearly every single occasion where it matters.

“Although we cannot say for certain that fixated vehicles have been processed, these findings do suggest that drivers’ fixation durations on these motorcycles were sufficiently long for them to be fully processed on other occasions… Previous theories regarding attention and awareness have suggested that awareness will occur when a sufficient amount of attention is allocated to an object, therefore a longer fixation, from which we are inferring more attention, would increase the likelihood that the object will be consciously perceived. If this were the case, it would have been expected that more frequent and longer fixations would be associated with reported, as opposed to unreported objects. We do not find significant evidence for this, with no large differences in the number of, or length of fixations on reported and unreported vehicles.”

That’s more or less in line with my thinking that it’s not so much a failure in the way that drivers search for motorcycles, it’s a failure in the way the brain processes the information.

What might cause that failure? This section began with a discussion of the problem of ‘workload’ and the finite limits of the human brain when there is a lot of information to be processed. The authors noticed that:

“…when drivers failed to report a motorcycle they had made more head movements and waited longer before pulling out after the initial fixation than on occasions where they reported it.”

In their discussion of their results, they state:

“While the explanation of a failure in visual attention may account for at least some report failures, we must also consider the possibility that some of these errors may occur due to a failure in visual working memory. For the current findings, this explanation is particularly compelling as the results suggest that it is drivers’ subsequent behaviour which predicts their ability to report vehicles. For this interpretation, information held in visual working memory may be subject to interference by subsequent visual information. Head movements in this situation will provide a LARGE QUANTITY OF NEW VISUAL INFORMATION TO PROCESS AND RETAIN, and there is limited reason to believe that subsequent visual behaviour after fixating on the vehicle would predict earlier attentional errors.” [my capitals]

And they also flagged up previous research which has investigated drivers’ memory by testing ‘working memory load’ by changing the number of visible vehicles from three and eight vehicles:

“It was found that the percentage of vehicles recalled decreased with increases in memory load, with drivers on average, recalling five vehicles when there were eight vehicles present.”

In other words, in an environment where there are few vehicles – and assuming that the bike is where it CAN be seen – drivers (and other bikers incidentally) WILL recall the PTW as they plan their manoeuvre. But in a crowded environment, where there are many more vehicles, cyclists and pedestrians to keep track of, there is a risk that a motorcycle, although it WAS seen, is then forgotten as newly-detected objects catch the driver’s attention and fill our short term ‘micro-memory’ with newer data. It would seem to be analogous to computer ‘first in, first out (FIFO)’ memory. And the number of discrete pieces of data – the buffer size – is small.

So to help understand Looked But Failed To See errors, the authors have created an entirely “new model of dynamic risky decision making in which the role of short-term memory is emphasised. The model, the Perceive, Retain, Choose (PRC) model, expands on those previously used to provide a much more explicit series of cognitive processing steps that may be involved in the decision to pull out at a junction or make other risky dynamic decisions.”

The diagram illustrates:

“…five potential pathways for relevant information to be used in the decision to pull out at a junction.

“Pathway 1 is the traditional account in which we look at the scene, encode the visuo-spatial information in it, and use this directly to decide whether it is safe to pull out. There are some situations where this information is all that is required, however, more commonly it will be necessary to combine visual information from one head movement with that acquired from later head movements.

“Pathway 2 thus involves the retention of information from the first head movement in visuo-spatial working memory. We assume that such information is retained in a limited capacity store and is available to a central processor at the same time as new information is being acquired. Clearly anything that interferes with the retention of such information will allow people to make accidental unsafe decisions.

“It is worth noting here that a common pattern of head movements in our studies involves a head movement towards a motorcyclist, then one to a car coming from the other direction, and a final one on the road ahead before pulling out. This raises the possibility that information from the second or third head movement has overwritten the initial contents of visuospatial memory, and that these were no longer available at the time a decision to pull out was made.”

The authors conclude:

“Our results suggest that some junction crashes in which a driver reports being careful and attentive in their visual checks but nonetheless pulls out in front of an oncoming motorcycle, could be misclassified. While previous researchers suggest that this crash is associated with a failure in drivers’ visual search for motorcycles – ‘Look But Fail To See’, the current results highlight the possibility that at least some of these crashes could occur due to a memory deficit – a ‘Saw but Forgot’ error.”

Final point from me.

This pair of papers would seem to fill in a crucial hole in our understanding of junction collisions. The first shows just how driver ‘workload issues’ can lead to focusing on a limited part of the environment. The second explains how motorcycles in plain sight could be seen but forgotten again thanks to the need to process visual inputs beyond the processing limits of the human brain – what I’ve called ‘micro-memory lapses’.

Either or both can cause the breakdown of the seemingly ‘simple’ task of “looking for a motorcycle”.

So next time someone says that “drivers should look harder for bikes”, just have a think about how complex the task actually is in terms of what’s going on in our brains. The surprise is not that drivers fail to spot motorcycles (and other vehicles) but the fact that they spot them many, many more times than they don’t. And those limitations will ALWAYS result in errors so long as humans pilot vehicles manually.

——————————–
NOW AVAILABLE – LIVE SOBS PRESENTATIONS
CHECK OUT SURVIVAL SKILLS ONLINE TALK SERIES – NEW!
https://survivalskills.tidyhq.com/schedule/events
——————————–

Since you’re here, I’ve a small favour to ask. If you feel able to make a small donation to the upkeep and continued development of SOBS, why not buy me a coffee? Each contribution is much appreciated. Each cuppa keeps me awake and writing! Thank you.

Ko-fi_Red

Continue your journey to the next page…

Last updated:

Friday 1 October 2021 – major edit to include the looked, saw and forgot ‘micro-memory’ failure issue
Friday 23 November 2018 – minor edit for clarity

References:

Helman, S., Weare, A., Palmer, M., Fernandez-Medina, K. (2012). “Literature review of interventions to improve the conspicuity of motorcyclists and help avoid ‘looked but failed to see’ accidents”, Published Project Report PPR638, Transport Research Laboratory Road Safety Group.

Murphy, G., Greene, C., M. 2016 “Perceptual Load Induces Inattentional Blindness in Drivers”

Pammer, K., Sabadas, S., Lentern, S. (2017) “Allocating Attention to Detect Motorcycles: The Role of Inattentional Blindness. Human Factors”: The Journal of the Human Factors and Ergonomics Society

Chloe J. Robbins, Harriet A. Allen, Karl A. Miller, Peter Chapman (2019) “The ‘Saw but Forgot’ error: A role for short-term memory failures in understanding junction crashes?”: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0222905

IMPORTANT:

The material is free to all to access and use under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. That means you can share it with your family and friends, and re-use it for club magazines and websites, so long as you acknowledge the source and author and include the same Creative Commons license in the derived works.

Please note, this Creative Commons license excludes commercial use. If you wish to use any of my work for commercial purposes, including (but not limited to) articles in pay-for magazines or commercial websites, please contact me.

Creative Commons statement

Kevin Williams has asserted his right under the Copyright, Designs and Patents Act, 1988, to be identified as the author of this work.

Photo credit Paul Townsend https://www.flickr.com/photos/brizzlebornandbred/20001313491