2.3 Pedestrian Crowds
2.3.1 Path Planning Methodology and Interaction
The manner in which a pedestrian behaves and interacts with others and their cur- rent environment, depends on factors concerning [55, 160]: the environment (present and future); population density; gender; and personal preferences of individuals. Such a large number of contributing factors makes it difficult to successfully predict how an individual will behave. For example, certain environments or age groups will express more likely behaviours, whilst the customs and traditions of individuals will potentially limit the number of possible outcomes [160]. It is however difficult to predict how everyone will react in a given situation as pedestrians exhibit free will. Despite being able to potentially estimate movement through these stereotypes, there are many more parameters concerning the complex nature of the individ- ual, including [55]: will; communication; fitness; geographical knowledge, many of which are conflicting. Pedestrian route choice and walking processes are mostly subconscious, however it can be assumed that the route choice is based on utility optimization [12] with the desirability of a route dependent upon factors such as:
f1 Distance or travel time between origin and destination, which is applied to all
f2 Proximity of obstacles or other physical obstructions; closeness to walls, which
is applied to basic maps, such as static environments
f3 Stimulation of environment, and attractiveness (e.g. ambience conditions,
shopping windows, shelter in case of poor weather conditions). Factors that are determined by the individual pedestrian preferences
f4 Number of sharp turns and rapid directional changes (route directness), which
is applied to path planning optimisation
f5 Expected number of interactions with other pedestrians (level-of-service), rel-
evant to a pedestrian only environment
Empirical studies have shown these are not mutually consistent, whilst their im- portance will vary between each pedestrian. To facilitate the development of the AMR, these factors can be reduced to consider a pedestrian only environment. The walking behaviours and interactions of individual pedestrians within a crowd can be assumed and summarised as the desire to achieve the following, for Item f1:
g1 Arrive at their destination with a certain time (e.g. [161]) g2 Move at a constant preferred speed as they walk (e.g. [14, 162])
and for Item f5:
h1 Avoid collisions with other pedestrians (e.g. [73, 163])
h2 Find the easiest route through the crowd to their destination (one that involves
the least potential collisions) (e.g. [162, 164])
These points are the cornerstone of a pedestrian’s navigation in a pedestrian-only environment [165], with the other points removed for the sake of simplifying the number of unknown variables.
2.3.1.1 Proxemics
The most desirable trait would be for a pedestrian to avoid any direct collisions, as well as to a lesser extent prevent violating the personal space of others [23]. Table 2.4 provides an example of the commonly desired spatial ranges that pedes- trians prefer to remain between each other, dependent upon their relationships. For a scenario that involves walking in a public environment it can be safely assumed that the distance between individuals will be within a minimum of the "personal" zone, as "close" and "intimate" zones are reserved primarily for stationary embraces. However, groups of acquaintances moving together often walk within these proxim- ities [43]. In order to best avoid collisions, violation of the ‘personal’ zone should be avoided, which should also help prevent collisions with groups. Any gap between two pedestrian of less than the ‘personal’ zone may mean cutting through a couple
2.3 Pedestrian Crowds 41
Spatial Zone Range Situation
Close Intimate 0 - 0.15m Lover or closefriend touching Intimate Zone 0.15m - 0.45m Lover or closefriend only Personal Zone 0.45m - 1.2m Conversationbetween friends Social Zone 1.2m - 3.6m Conversation tonon-friends
Public Zone 3.6m+ PublicSpeaking
Table 2.4 Proxemics for human-human personal spatial zones [22, 23]. The desirable zones regarding the proxim- ity of one human to another. Study carried out on mainly urban English-speaking parts of the world.
Fig. 2.8 Elliptical proxemics distribu- tion around a human, elongated in front of
them (Reproduced
from [11]).
of friends, but also will create a higher collision potential on both sides of the pedes- trian cutting through. For a moving pedestrian proxemics are often represented with an elliptical distribution, Fig. 2.8, due to their current velocity. This is observed in empirical studies discussed later in Section 2.3.2.4.
2.3.1.2 Navigation
Crowds can be unpredictable due to the heuristic methods pedestrian use to plan their paths, which are excellent for complicated scenarios as it aims to improve strategies in order to create better decisions [166]. Humans strategically plan a route to their ultimate goal, but navigate along this route using a heuristic approach [167]. The "Categorization by Elimination" heuristic [168] uses successive visual cues to reduce the set of possible categories objects may belong to until only a single category remains, so that only necessary information is processed (e.g. when a pedestrian walks along a pavement the heuristic eliminates everything except what is infront of them on the pavement. For instance, cars on the road are of no concern until the pedestrian desires to cross it, in which case heuristic will eliminate everything except the cars passing and the distance required to cross the road. Once the pedestrian has crossed, the heuristic eliminates everything except the upcoming pavement, as before. This highlights the fact when undertaking a specific task, a human will be very selective on what information to process, using only the minimal information required. Complex behaviour can occur when responding to objects based on only a few simple rules, which has been previously utilised for crowd modelling [169]. How a pedestrian chooses to behave when path planning is split into three levels [12]:
Fig. 2.9 Visualisation of the pedestrian path planning choices. Partially derived from the text in [12], and inspired by the basic framework for applying a social force model found in [13]. The diagram shows the three behavioural levels of how a pedestrian plans their path, along with the three dimensional levels of how a human operates. Within the "Dimensional Level" the "Action" level taken by pedestrians is the same as the AMRs’ general autonomy cycle, Fig. 2.1b. Also, the "Spatial and Temporal" level employs the same navigation strategies as an AMR, Section 2.2.
i1 Strategic level: Departure time choice, and activity pattern choice.
i2 Tactical level: Activity scheduling, activity area choice, and route-choice to
reach activity areas.
i3 Operational level: Walking behaviour.
Fig. 2.9 highlights the modelling framework for pedestrian behaviour when path planning. Similar to robotic path planning [96] both global and local path plan- ning is used. The "strategic" level (Item i1) involves deliberate global path plan-
ning that finds the most desirable paths that go from start to finish. The "tac- tical" level (Item i2) involves selecting the best path en route that is influenced
by constantly acquired knowledge, updating the global path. The "operational" level (Item i3) involves instinctive local collision avoidance, which is the immediate
response to something unpredictable in the immediate vicinity.
As mentioned in Section 2.3.1, movement and path planning is dependent upon physical issues of crowd density, groups, and traffic direction. When focussing on path planning and collision avoidance for pedestrians in a dense urban environment, the most helpful and significant factors are ones that involve mutual interaction, especially when trajectories intersect. To establish the significance of this particular pedestrian interaction and collision avoidance scenario a number of empirical studies are analysed, along with an evaluation of any associated model attributed to the findings.
2.3 Pedestrian Crowds 43