Adaptive Decision-Making in Response to Crowd Formations in buildingEXODUS

Given the importance of occupant behaviour on evacuation efficiency, a new behavioural feature is under investigation for implementation into the buildingEXODUS evacuation model.  This feature concerns the response of occupants to exit selection and re-direction. This behaviour is not simply pre-determined by the user as part of the initialisation process, but involves the simulated occupant taking decisions based on their previous experiences and the information available to them. This information concerns the occupants prior knowledge of the enclosure and line-of-sight information concerning queues at neighbouring exits.

The manner in which occupants behave in response to crowd formations is of fundamental importance to the success of an evacuation.  Their ability to ascertain the likelihood of extensive delays and possibly alter their exit route accordingly is essential to the navigation process.  Occupants determine their choice of exit through examining a number of factors. Initially, the occupant must be aware of the existence of an exit. Obviously, inherent in this knowledge is the geometric layout of the enclosure surrounding that exit.  Occupant familiarity has long been seen as of fundamental importance to the progress of an evacuation. Instead of occupant’s heading towards the nearest exit – of which they may have no prior knowledge - as would be assumed by the majority of present building regulations, they are more likely to move towards other more distant exits with which they have had previous experience and with which they feel more confident.

The impact of familiarity upon the behaviour of the occupant is not limited to exit usage. Horiuchi recorded the increased levels of confidence which familiarity bred in occupants, allowing them to perform actions not directly linked with speedy evacuation. Familiarity with the enclosure may generate a level of confidence that allows the occupant to attempt activities such as fire-fighting, delaying their response or attempting to follow alternative, less direct routes. Although in the short term these routes may not be considered optimal, they will have been adopted through calculation on the occupant’s part to guarantee safe egress and to minimise the imminent risk and the evacuation time.  It is unlikely that occupants make decisions concerning redirection in isolation, but instead weigh up the data available to arrive at a final decision.  Influential factors which are likely to affect this decision include the length of the queue at any exit, the existence and severity of smoke and the distance to the exit. The movement of the queue will in itself be dependent upon the geometry (terrain, size of exit, etc.) as well as the make-up of the crowd. As a survivor of the Beverly Hills Supper Club incident recalls,

In this incident, before the occupant commits to another course of action, he is considering information concerning the exit configuration and the crowd conditions around each of the exits. Except for the occupant’s previous familiarity with the enclosure, all of the factors require the occupant to be in visual contact with the necessary environmental cues to make these determinations. Therefore, the decision to redirect egress movement is not solely based on factors determined prior to the evacuation, but is likely to be influenced by dynamic factors such as population size and environmental considerations.

The visibility of the exit determines the level of information that the occupant may use in any calculation of the tenability of any future use. For a thorough appraisal to take place, the occupant has to be in visual contact with the exit, to examine the surrounding population, environmental conditions, etc. If the exit is not within visual range, the occupant has to rely solely on his recollection of exit details from memory, such as position and distance, or possibly from information communicated to them from the surrounding population or from a procedural influence such as an intelligent alarm systems.

Finally, through examining these factors and their own experience, the occupant must come to a decision on a course of action. This might involve a crude determination of which route would enable the most ‘efficient’ and safest path of egress. As highlighted previously, this calculation can only be made in respect to the information available to the occupant and any previous experience that he might have. This represents the occupant as being capable of information processing as described in recent psychology literature.

The familiarity concept is introduced into EXODUS through the introduction of the DOOR VECTOR. This is a list of exits known to the occupant.

Occupants are also provided with ‘line-of-sight’ information concerning neighbouring exits. A complete ‘line-of-sight’ system would be computationally expensive and complex to implement. An alternative method has been developed to represent the ability of occupants to examine exits within their line-of-sight and reflect upon the information gleaned. In this simplified system, the user supplies the visibility status of exits. This is achieved through grouping the exits according to which of them can be seen simultaneously.

The proposed adaptive queuing behaviour is reliant upon the introduction of the door vector and the exit line-of-sight feature. Initially, the occupants’ situation is examined to determine whether he desires (i.e. estimated exit time is reduced) and whether it is possible for him to alter his target. This involves the examination of a number of factors including:

- the distance between the occupant and the exit,
- the extent of time occupant has spent waiting,
- the occupants patience level,
- whether occupant is completely surrounded by other occupants,
- the estimated time of arrival at the alternative exit,
- the estimated time of arrival at the current exit, etc.

As an example of how these factors influence an occupants decisions, consider the following case.  An evacuating occupant has waited in an exit queue for a period of time greater than his patience level. He is also sufficiently distant from the exit not to be committed to using the exit. He is therefore now willing to consider redirection. The occupant is situated on the periphery of the crowd and is therefore able to contemplate redirection.  Examining the exits available to him, he determines that he will arrive at another exit more quickly. The occupant therefore, moves off towards his new target.  This decision making process contains a stochastic element and will therefore alter between repeated simulation runs.

There is of course no guarantee that recommitting to another exit will produce a better outcome for the individual (i.e. decrease personal evacuation time).  Taking this course of action may result in a sub-optimal outcome for the individual concerned.  Furthermore, by allowing the occupant to move to an unseen exit, the chances of the occupant delaying his evacuation are greatly increased.  This is due to the fact that the unseen exit may in-fact not be viable due to the extent of crowding around the exit. As the exit is unseen, the occupant is deprived of this information and essentially takes a chance.

In the prototype implementation of this behaviour, the simulation is shown to provide a more complex and arguably more realistic representation of human behaviour than that provided by the existing model. Furthermore, the implementation demonstrates the significance to both the evacuation as a whole and the occupant as individuals of the inclusion of such behaviour.

If the occupant is able to utilise his ability to determine a more effective route through the analysis of exit crowding, then the optimality of the evacuation is increased. However, the capability of the occupant to switch between available exits does not guarantee the reduction of individual and total evacuation times.

For more information about evacuation modelling and the EXODUS software visit the EXODUS Web Pages. For a complete listing of EXODUS and evacuation publications visit the FSEG Publications pages.