CHAPTER 2 DRIVING AND DRIVER BEHAVIOUR MODELS
2.5 Driving Behaviour Models
2.5.1 Performance Models (Behavioural Task Analysis Models)
Task Analysis Models describe specific goal-directed human activities and are data driven. They describe the task according to the task requirements (facts about driving), the performance objectives (behavioural requirements) and the enabling objectives (aptitude requirements) that are needed in order to perform the task successfully.
These models of driving behaviour depict driving as being part of a physical system. They are concerned with how drivers perform specific tasks and what causes them to make errors (Rothengatter, 2001). To understand how specific tasks are performed, they are broken down into a series of less complex tasks (Kirwan & Ainsworth, 1992; Robertson & Thoreau, 2003). The chronological order of these tasks may change depending on the circumstances the driver is faced with. McKnight and Adams (1970ab) formulated a Task Analysis Model of driving by partitioning driving into forty-five major tasks. These tasks were composed of 1700 elementary tasks. The enabling objectives that were required in order to perform the driving task successfully were knowledge and skills.
According to Robertson and Thoreau (2003), Task Analysis Models do not only have to take into account recommendations from the Highway Code they can also incorporate informal road rules. In a study exploring pedestrian and driver behaviour whilst crossing roads, Robertson and Thoreau (2003) incorporated tasks into their Task Analysis Model, which have become part of the way people use road crossings (Figure 2.3). For example, some drivers wave at pedestrians to signal to them to cross the road, or may flash their headlights at other cars to signal them to go ahead. These actions, whilst not recommended by the Highway Code, have become unofficially accepted as a norm through common usage. Task Analysis Models can therefore consist of both formal and informal road rules.
Task Analysis Models of driver behaviour are one of the main driving forces behind the development of computer simulation in traffic analysis as they continue to evolve and become more sophisticated. For more than four decades computer simulation has aided research, planning, demonstration and the development of traffic systems (Pursula, 1999). Task analysis models have been influential in the design of driver training simulators and in-vehicle devices such as adaptive cruise control (ACC), automatic warning systems (AWS) and automatic braking systems (ABS).
2.5.2 Servo-Control and Information Flow-Control Models
The Adaptive Control Models describe driving as either Servo-Control Models (a set of continuous or intermittent tasks) or Information Flow-Control Models (a set of flow charts or decision trees).
Servo-Control Models represent skills involving steering or obstacle avoidance. They have been particularly important in understanding the interaction between the driver and the vehicle, for instance they identify how drivers‟ react and respond to cues from the external environment and from the input signals experienced from the vehicle. These models act on input signals represented by cues from the lateral position of the vehicle on the road, compensatory tracking or cues from the visual scene about the roads geometry and pursuit tracking. Drivers‟ anticipation and slowness to react are accounted for by lead and lag components. Klein, Vincent and Isaacson (2001) found that experienced drivers are more attentive to environmental cues than novices who tend to use heuristics and basic signals.
Figu re 2.3 – An E xampl e of a T ask Anal ysis M od el for a Dr ive r E n cou n te rin g an d Ne gotiat in g a Cr ossi n g (R ob er tson & T h or eau , 2003)
The Threaded Cognition Model (TCM) proposed by Salvucci and Taatgen (2008) is a contemporary servo-control model. Threaded Cognition is an integrated theory concerned specifically with non-deliberative concurrent multitasking performed at the sub-second to second time scale (i.e., performing more than one task simultaneously, such as driving and dialling a number on a mobile phone). This computational model was designed to understand, model and predict performance during concurrent arbitrary tasks. It can also be used to explain how multitasking behaviour can result in interference. This model has been praised for its ability to test predictions.
Salvucci and Taatgen‟s theory proposes that streams of thought are represented as threads of processing. Complex dynamic tasks like driving incorporate multiple task threads (Salvucci & Taatgen, 2008).
According to the TCM, information threads are coordinated by a serial procedural resource. This procedural resource is employed to do several things:
1) to allow the concurrent execution of the threads across available resources (e.g., motor and perceptual resources),
2) to acquire resources and, 3) to resolve conflicts.
Salvucci and Taatgen‟s model proposes that multiple tasks can be processed in parallel. Threads acquire resources in a greedy manner by requesting resources as soon as possible. Conversely, when the resource is no longer needed the threads release them politely. When a thread requires a specific resource that is busy, it waits until the completion of the current process before acquiring the resource. Two or more threads may have to wait to acquire the same resource.
Resource acquisition can only occur through rule firing. The least recently processed thread (i.e., the thread which has not recently fired a rule on the procedural resource) is allowed to proceed first. This provides a parsimonious balance between threads. Conflicts arise when tasks require the same peripheral resource or when multiple tasks require attention from the central procedural resource. Conflicts for resources reduce parallelism and lead to processing delays. When two tasks require common perceptual or motor resources dual-task performance for one or both tasks will be impaired (Salvucci & Taatgen, 2008).
There are four cognitive resources that are required in the Threaded Cognition model which are modules from Anderson‟s ACT-R (Adaptive Control of Thought-Rational) architecture (Anderson, 2007). According to Anderson each task that humans perform consists of a series of discrete cognitive and perceptual operations. The four resources in the Threaded Cognition Model are as follows (Salvucci & Taatgen, 2008; Taatgen, Juvina, Schipper, Borst & Martens, 2009):
1) Visual module – to perceive items (input)
2) Procedural memory – where conditions from the other modules are mapped into actions. Information is integrated here and, with practice, the task instructions that were encoded as chunks in declarative memory can be changed into production rules that can affect new behaviour. Production rules can discriminate when a particular resource is in use (i.e., when the module is busy or the buffer is full).
3) Declarative memory – determines whether items are targets or distracters and stores factual knowledge in chunks that can be recalled or forgotten. Requests to retrieve information chunks based on partial patterns can be processed here one-at-a-time.
4) Imaginal module – a limited working memory store that is important in memory consolidation
According to Salvucci and Taatgen (2008), driving requires the repeated firing of four rules that iterate in sequence. Each of these rules provides updates for adjustments that need to be made to steering and acceleration. These four rules are as follows:
1) Find the near point.
2) Find the far point of the current lane (information about nearby and upcoming lane configurations to help calculate steering angle).
3) Motor commands are sent to specialised motor modules for steering and pedal movements and also directs visual attention to encode the information at the far point (i.e., a road point or lead vehicle).
4) Check for the stability of the vehicle by monitoring the vehicle‟s lateral position and velocity – if stable this process iterates after some delay.
To illustrate how the driving rules operate and the impact of performing a secondary task, Salvucci (2001) conducted a study that looked at the impact of dialling a number on a mobile phone whilst driving (Figure 2.4). Salvucci found that when two tasks were performed simultaneously which required different resources they could be performed successfully (for example, noting that the vehicle is stable requires the procedural resource and retrieving
telephone digits requires the declarative resource). However, if the two tasks required the same resource they would need to compete for it (for example, finding the road near point and retrieving blocks of numbers both require the procedural resource). In this situation only one of the tasks would successfully acquire the resource whilst the other would have to wait until the resource was given up (in the aforementioned example, when the driving task has won the resource the retrieval task must wait and vice versa). In a similar study, Salvucci and Macuga (2002) concluded that due to competing for resources, performing the dialling task had a significant potential to result in driver distraction and ultimately decreased performance on the primary driving task.
Figure 2.4 – Driving-dialling study: Model timeline (Salvucci 2001, cited in Salvucci & Taatgen, 2008)
Information Flow-Control Models use digital computer simulations in their attempt to simulate driver behaviour. The Kidd and Laughery (1964) information flow model is a dynamic Task Analysis Model which specifies that when particular conditions are fulfilled certain acts will follow. For example, drivers receive vital visual information as they drive
along which needs to be processed. The angles of approaching objects, such as other vehicles, need to be checked and the points of intersection calculated. If the visual angle remains constant (i.e., the distances from vehicles/objects to the point of intersection remains the same), a collision is imminent and the driver cannot take any evasive actions.
Michon (1985) criticised the Kidd and Laughery (1964) Information Flow-Control Model for being too data driven and for having little to do with cognitive modelling. Once parameters are determined the program runs on fixed algorithms and there is no room for either intelligence or learning in the model. In the presence of pedestrians the model would not stop to allow them to cross the road, instead it would continue moving and thus run them over. This is the result of the inflexibility of the program, which does not have any real priority interrupts and is too rigid in its approach.
The Task Analysis Models, Information Flow-Control Models and Servo-Control Models are useful for furthering our understanding of how people drive (from a cognitive perspective) and why errors are made. However, they are unable to take into account the effects of individual differences (e.g. personality, age, gender) and motivations (e.g. choosing to drive fast to get to a meeting on time) on driving styles.