Behaviour is a result of the interaction betweenneeds,actionsandemotions. To model this interaction, the correlations betweenneeds,actionsandemotionsare first modelled.
From these correlations, the determination of theactionsto be performed, is described, based on these correlations. Decision making is shown as the result of the interactions between these concepts.
4.8.1 Correlation between needs, actions and emotions
The purpose of correlations is to represent relationships between the concepts of emotions, actions, and actions. Such relationships may be causal, or bi-directional.
These correlations, as set by the system designer, forms a part of the design process.
Such correlations are used in decision making.
Need-action correlations
These are represented as a matrix of the following form:
Cnα=
c1,1 c1,2 . . . c1,M
c2,1 ... c2,M
... ... ...
c|Ar|,1 c|Ar|,2 . . . c|Ar|,M
; Cnα=CnαT (4.57)
where M is the number of needs, and |Ar| is the number of actions, and each value ci,j ∈[−1,1]∩R. This matrix represents the relationship between the intensity of each need, and the motivation for eachaction.
Action-need fulfilment correlations
The correlations betweenactionsandneeds fulfilmentare represented as:
Cαn¯ =
d1,1 d1,2 . . . d1,|Ar|
d2,1 ... d2,|Ar| ... ... ... dM,1 dM,2 . . . dM,|Ar|
; C¯nα=Cαn¯
T (4.58)
From the understanding that the presence of needs will motivate actions and the execution of actions would lead to the fulfilment of the same needs, the correlations Cα¯ncan be computed as the negative ofCnαT, namely,ci,j=−dj,i, fori∈[0, M], j∈[0,|Ar|].
Need-emotion correlations
The correlation betweenneedsandemotionsare represented as in equation (4.52):
Cnε=
g1,1 g1,2 . . . g1,P
g2,1 ... g2,P
... ... ... gM,1 gM,2 . . . gM,P
; Cnε =CnεT (4.59)
whereM is the number ofneeds, andP is the number ofemotions.
Emotion-action correlations
The correlation betweenemotionsandactionsare represented as:
Cεα=
h1,1 h1,2 . . . h1,|Ar|
h2,1 ... h2,|Ar|
... ... ... hP,1 hP,2 . . . hP,|Ar|
; Cεα=Cεα
T (4.60)
where|Ar|is the number ofactions, andP is the number ofemotions.
Cross-correlations between all concepts
Representing theneeds, emotions andactionsas a vector V = [nT, ET, AT]T, the cross correlation matrix is given as:
CV=
I|M×M| Cnε Cαn
Cnε I|P×P| Cαε
Cnα Cεα I||Ar|×|Ar||
(4.61)
The correlations between needs, emotions and actions as represented in CV is to be designed by the system architect, to set the desired relationships between these concepts once they have been identified.
4.8.2 Selection of desired actions
Maximizing the likelihood that a system can operate autonomously over long periods of time, while ensuring its survival, is equivalent to an optimization problem that maximizes a fitness function i.e. equation (4.1):
QN :ak =arg max
a∈ar{Ffitness(nk)}, (4.62)
where Ffitness is a fitness function. QN determines the action ak that maximizes the needs fulfilmentof the robot, given the currentneedslevel nk. This involves finding a trajectory in action spaceArthat leads to a value ofnk whose elements are minimized (i.e maximize fulfilment). A possible candidate forFfitness isk nˆ¯k+1 k, where nˆ¯ is an estimate/prediction of theneeds fulfilment vectorin the next time instant. Other fitness functions can be designed based on the different survivability metrics as discussed in section B.1.
Despite this formulation, it may be sufficient fornk to besatisficedto some level of acceptability n¯∗ = [¯n∗1, ¯n∗2, ..., n¯∗M]T for survival in the same manner depicted in the survivability specification, as opposed to determining the optimal solution. The concept of satisficingin decision making is based on bounded rationality theory (Simon, 1957)
Current Needs Assessment
Mobile Platform
(Vehicle) Environment
Perception World Model
Knowledge Representation Current
Needs Level
Emotional Assessment Desired
Needs Level
Needs Deficiency
Emotional States
Behaviour Selection
Selected Behaviour Desired Needs
Assessment Goals
+ -
Percepts System States
Figure 4.2: An illustration of the process ofneeds-fulfilment, in the form of a closed-loop control system. The actions or behaviour of a system are determined by the deficiency in needs, and theemotional statesresulting from the level ofneeds-fulfilment.
and refers to the process of determining solutions to problems that cansatisfycertain aspiration levels (i.e. criteria) to an extent that is sufficient for achieving a system’s purposes. It is usually used to describe how humans do not necessarily search for optimal solutions to problems, but settles on those which are good enough. Optimality comes at a price if the extra resources or effort taken to find it is too great, especially if the existing solution already suffices. In the present context of action-selection, obtaining optimal solutions may be intractable given incomplete knowledge ofAr and n˙r = F(nr,ar), or the lack of the means to evaluate the optimality ofak especially in a complex system.
In reality, designed-actions and behaviours are not simple input-output systems, but themselves be complex or dynamic systems in their own right e.g. exploration, mapping of environment, surveillance, intrusion detection. Hence it is not always possible to express each and every action inAras a simple mathematical expression or function.
4.8.3 Needs fulfilment as a closed loop system
To illustrate the concept of needs-based control, a diagram showing the translation from needs to behaviours is shown in Figure 4.2. From the goals of the autonomous vehicle, a set of desired needsfulfilment levels is generated. The current needsof the system is generated from the internal states of the system. Emotion-based assessment of the difference between the currentneedslevel and the desiredneedslevel results in emotional states that assist in the selection of appropriate behaviour for execution.
4.8.4 Managing unanticipated scenarios
No assumptions are made so far about the observability of the system. There could be a number of unobservable system states which correspond tohidden needs. To overcome
such a constraint, the design of an observer is relevant, and this would function as a process ofneeds-estimation.
Handling unanticipated scenarios is a major requirement in autonomous robotic systems operating beyond a “shoe-box” environment. Various techniques such as case- base-reasoning, nearest-neighbour interpolation, and artificial neural networks have been applied in solving such problems. They focus on the determination of a best- case response to a novel situation, which entails the ability to recognize the lack of an instance within the experience in responding to the novel situation. However, the identification of a novel situation from typical operating scenarios is a difficult task, as it requires a comparison with previously encountered events to determine if the deviation from these events warrants a treatment of the situation as a new event.
Withneedsassessment, this problem can be circumvented. Since decisions on “what to do next?” are based on the relative degrees of fulfilment of the needs of a system, unanticipated scenarios need not be explicitly detected; if theneedsof the system have been modelled in a comprehensive manner, the presence of a fresh scenario is reflected as changes to theneeds fulfilment. For example, a robot which is encountering slippery terrain for the first time may react to it by observing the terrain or travelling at slow speed until it is able to determine that it is capable of traversing it safely.