UK UNCLASSIFIED

A new technique to address CID and IFF studies

David Dean, Kathryn Hynd, Beejal Mistry, Alasdair Vincent and Paul Syms

Dstl Information Management (IMD) and Land Systems Departments (LSD),

Portsdown West and FortHalstead

Key words: Amicide, combat identification, Dempster-Shafer method, fratricide, friendly fire, human factors, stochastic simulation

Abstract

Effective combat identification (CID) of entities encountered in the battlespace is critical. When CID fails, the consequences of fratricide (‘friendly fire’) and missed opportunity can impact on mission effectiveness, morale, and lead to political repercussions. The MoD believes that there are three broad approaches to improving CID: situational awareness (SA), tactics, techniques and procedures (TTPs), and identification friend or foe (IFF) devices.

This paper describes the Integrative Combat ID Entity Relationship (INCIDER) model, which was developed to compare the relative effectivenesses of these interventions across all three approaches. INCIDER combines a simple represent-ation of the physical world as seen by a single tactical decision-maker (DM, e.g. a tank commander) with a simple model of the DM’s cognitive processes when confronted with an unknown target entity. The physical model represents the sensors, communications and IFF devices on the DM’s platform, together with the weather and target characteristics, and allows the platform to move (e.g. to close the range to the target). The cognitive model combines sensory, SA, IFF and other information with the DM’s confidence in these sources, using the Dempster-Shafer method, and also represents the effects of preconceptions and confirmatory bias. Both models can represent the stresses of battle. The physical model feeds the cognitive with information, and the cognitive model decides on what further action the DM will take. When a threshold is reached, a stochastic simulation is used to determine the CID decision that they take.

INCIDER is a significant development in the assessment of the contribution of human factors in warfare, since it is able to combine and compare physical and psychological effects quantitatively, using the same technique.

Copyright statement

 Crown copyright 2005, Defence Science and Technology Laboratory [Dstl]

Author for correspondence:

Dr. P.R. Syms, Bldg. A20, Dstl FortHalstead, Sevenoaks, Kent TN14 7BP UK

Tel.:+44 (0)1959 892086

Fax:+44 (0)1959 892504

E-mail:

Introduction

Historically, amicide (fratricide, Blue-on-Blue, or friendly fire) has been a feature of warfare since the beginning of recorded history. During WW1 and WW2, it is thought to have accounted for 10–20% of the overall casualties. In the recent conflicts in the Gulf, the mismatch between forces led to low overall numbers of US and UK casualties, highlighting those caused by our own forces. The public is naturally reluctant to accept any casualties in war, but when these casualties arise from misidentification, the political impact is increased.

Effective combat identification (CID) of entities encountered in the battlespace is critical in reducing amicide. CID is defined by the British Army as:

‘The process of combining situational awareness, target identification, specific tactics, techniques and procedures to increase operational effectiveness of weapon systems and reduce the incidence of casualties caused by friendly fire’

Good combat identification can improve the tempo of operations, and increase the effectiveness of manoeuvre and engagement. Conversely, poor CID can:

  • lower tempo (wasting time as unknown forces are positively identified);
  • introduce Blue fratricide casualties, lowering morale and effectiveness;
  • incorrectly cause civilians to be targeted;
  • waste resources to engage or guard against non-existent threats;
  • expose own forces to danger from incorrectly identified hostiles;
  • lead to political repercussions, both nationally and within coalitions.

The MoD believes that there are three broad approaches to improving CID:

  • better situational awareness (SA);
  • physical devices, e.g. identification, friend or foe (IFF);
  • training in tactics, techniques and procedures (TTPs).

Assessment of CID is not easy. It involves elements of the physical domain, such as military equipments (e.g. vehicles, uniforms, sensors etc.) and the environment (e.g. terrain and weather), together with elements of the informational (plans, briefings and communications), cognitive (e.g. training, and resolving conflicting information) and psychological (e.g. character, expectation, stress, fatigue) domains.

Traditionally, the physical domain has been studied using wargames and combat simulations, the informational using process models (sometimes in alliance with combat models), whereas the cognitive domain was studied separately, and approached much more qualitatively using ‘soft OA’ techniques. But for MoD to assess quantitatively the cost-effectiveness of CID solutions that spanned all these domains, it was necessary to develop a single quantitative method.

Moreover, the MoD requirement was for a model that could represent decision-makers (DMs) in the sea, land and air environments, on joint (bi- and tri-service) and combined (coalition) operations. In 2004 Dstl judged that the problem across these domains was understood sufficiently well for quantitative modelling to be attempted, and that sufficient data existed or could feasibly be gathered to support it.

The INCIDER model

To meet this need, Dstl have developed an holistic andconsistent set of parameters across all these domains, and a generic iterative, stochastic decision-making assessment tool called the Integrative Combat Identification Entity Relationship (INCIDER) model.

INCIDER integrates physical representations of sensors and IFF systems with human cognitive and behavioural characteristics, and can represent simplified detection and classification processes within an instantiated representation of an encounter set within an operational context. It also considers how operational characteristics will impact upon the implementation of the human and physical factors.

Figure 1. Summary of INCIDER inputs and outputs

INCIDER represents a single decision-maker observing a single ‘target’ entity on a flat battlespace, through a number of channels. TheDM can also refer to third party information sources, such as higher HQs, for more information on the identity of the target. The process is generic and can represent any type of DM or target entity within any environment.

After initial detection, the decision maker will undertake an iterative process to gain more information about the target by using the available information sources, and moving closer to the entity to improve the performance of his sensors. Eventually a confidence threshold level will be reached, at which time an assumed identity (correct or incorrect) will be assigned, or he will be ‘timed-out’.

INCIDER inputs: The model requires data on the sensors available to the DM, and look-up tables of their performances, based on existing physical and experimental models. It requires the ground truth target identity (Red, Blue, neutral or non-target[1]), and initial detection range. It also needs a wide variety of parameters that describe the human aspects of the decision making process, such as data on the observer’s expectations, personality (currently described using the Myers-Briggs system, which influences the propensities for different actions), training (e.g. skill in physical identification), stress and fatigue.

Preconception: A decision maker will enter an encounter with a preconception about the identity of the target. This will be based on previous history and the pre-mission briefing. Values relating to the relative belief in each of the identity options (Red, Blue, etc.) are key inputs to the process.

Decision threshold: There is a threshold level of evidence required to make a decision. This level will be determined by a mixture of the rules of engagement (RoEs), experience, fear or threat and the tactical situation, and personality. The decision threshold will tend to lower as the encounter develops, and potentially decrease rapidly with distance if the DM feels under threat from the target.

Confirmatory bias: All Individuals will tend to see what they expect to see. This is why preconceptions are so important in the decision making process. Moreover, individuals will tend to seek out information that supports their preconception, and reject information that contradicts it. This phenomenon is well recorded, but very hard to represent. Within INCIDER, the stronger the preconception belief, the higher the level of confidence in new information must be in order to be believed; lower grade information is ignored. This has been represented using a filter on the incoming information.

Pre-set human characteristics: These characteristics consider aspect of the DM’s experience and character type. Currently this includes:

Myers-Briggs personality indicators are used to determine sensory preferences, and the likelihood of using different types of information source (i.e. a bias towards human sources, technology or SA pictures).

Level of training represents the degree of proficiency that an individual has in identifying a target, used to moderate the information provided by imaging sensors.

Level of experience represents the amount of time that an individual has worked alongside colleagues, and hence will be more aware of their location through familiarity (and is used to modify preconception).

Figure 2. Details of the decision-making models in INCIDER

Previous fratricide: An individual who has been involved in a previous fratricide incident may tend to be more risk-averse, requiring a higher decision threshold.

Variable characteristics:

Stress indicates the degree to which cognitive resources are limited by other activities. Fatigue indicates the degree to which cognitive resources are limited by physical tiredness and lack of sleep. Both stress and fatigue will either cause people to polarise their thoughts, i.e. become more biased (going with their preconceptions, because it is easy) or refuse to make a decision (raising their decision threshold).

The encounter model: A ‘fusion engine’ uses a Bayesian-derived technique, the Dempster-Shafer method, to combine inputs from different sensors and SA tools to derive overall ‘confidence masses’ describing the DM’s belief in the target’s identity.

This level of confidence is fed into a ‘decision engine’that compares the history (expectation) confidence masses with the newly derived observation mass. The effect of confirmatory bias filters out weak inputs, and a modified set of beliefs results.

The decision engine also determines by a stochastic process the next action that the DM will take (e.g. to move closer, use a different sensor or SA tool, contact HQ, etc.). This is an iterative cycle where the DM closes range with the target (see Figure 2). The belief from the previous iteration becomes the new expectation, and the DM uses a mixture of information sources to raise confidence.

Eventually, when a particular confidence mass exceeds the decision threshold, a decision is made. INCIDER records the result of the decision, ant the time taken to reach it. It does not attempt to model the consequences of the action taken on this decision (e.g. an engagement and possible return fire).

Because INCIDER is stochastic, and outputs only one decision per run, the runs are replicated many types – typically 1000 times – so that the results across a range of treatments can be analysed statistically using the Analysis of Variance.

The way ahead

INCIDER has successfully been able to capture parameters relevant to combat identification, which have been used to generate an encounter model which represents human decision making, and has been regarded by military officers as the ‘best representation of what it is to be a tank commander’.

Figure 3. Validation of INCIDER, and its relationship with other CID modelling

To date, INCIDERhasonly been validated to date using military judgement. Dstl intends to incorporate it into a programme of CID experimentation (see Figure 3). The next stage, taking place later in 2005, will be to undertake synthetic environment (SE) experiments in collaboration with QinetiQ Ltd. to validate parameter behaviour and calibrate the encounter model. The SE in turn will be validated against a live exercise due to take place on Salisbury Plain Training Areain October 2005. Dstl will also be giving a number of psychometric tests to the participants of both the SE and the field exercise, in order to try to assess the range of personality and other psychological characteristics among the Army decision-maker population, and attempt to correlate these with decision-making behaviours. This may lead Dstl to adopt different psychometrics in the model, if different characteristics are shown to be important.

Initial runs showed ‘binary’ results in some situations, where the same decision was reached by 100% of DMs in a given run. We shall be re-visiting the model to improve the realism of the error and failure modes in the physical, informational and cognitive domains, so that it better reflects the ‘Swiss cheese’ nature of errors in safety-critical systems.

In the longer term, Dstl aims to improve the representation of CID that is incorporated into high level wargames and constructive simulations in order to support balance of investment analysis, and inform CID decision-making across all the MoD’s lines of development. The favoured combat model to experiment with this linkage is Dstl Land Systems Department’s Close Action Environment (CAEn).

Conclusions

INCIDER is a significant development in the assessment of the contribution of human factors in warfare, since it is able to combine and compare physical and cognitive domains quantitatively, in a single quantitative model.

The model has been able to demonstrate to required characteristics of human decision making, and once calibrated will enable the assessment of expected misidentification rates within a particular operational context. The decision-making process in INCIDER is generic, and has potential application to a wide variety of situations in whichcomplex decisions are made, and incomplete or contradictory evidence from a wide variety of sources is available.

1

UK UNCLASSIFIED

[1]Non-targets have been included because it is possible for wildlife, rock formations, broken down vehicles and derelict structures to be mistaken for military targets. In one example from the 1991 Gulf War, two hawks sitting on a water tank were mistaken for an enemy observation post, and indirectly led to an amicide incident.