See the article on our work in the New Scientist, 28th June 2003 and coverage from the Australian Broadcasting Company Sept 2003.

Motion camouflage is a stealth behaviour that allows a predator to conceal its apparent motion on approaching a moving prey. This page provides an overview of current motion camouflage research (Sept, 2003) including a description of work undertaken by myself and Peter McOwan concerning the computational modelling of the behaviour.

Motion camouflage, first described by Srinivasan & Davey (1995), is a stealth technique that allows a shadower (e.g. a predator) to approach a moving shadowee (e.g. the prey) whilst appearing to remain stationary. To achieve this, the shadower follows a path such that it always lies on the line connecting the shadower and fixed point (see Fig.1). If the shadower is a predator approaching its prey, ignoring the movement of limbs (etc) in biological organisms, the only visual cue to the shadower's approach is its looming (i.e. the increase in the shadower's image size as the distance between shadower and prey decreases). In contrast with any non-camouflaged approach, the prey perceives no lateral or vertical movement away from the direction of the fixed point. The fixed point could be an existing landmark in the background (against which the shadower is camouflaged) or possibly the initial position of the shadower (in which case the shadower appears not to have moved from its starting point).

Motion camouflage was initially inspired by observations of the mating tactics of male hoverlies tracking females (see Srinivasan & Davey, 1995). Upon spotting a female, the male hoverfly Syritta pipiens appears to shadow her, maintaining a constant distance until she lands, at which point the male swiftly moves in to mate. During the pursuits examined by Srinivasan & Davey (1995) (originally filmed by Collett & Land (1975,1978)), the males periodically appeared to switch fixed point locations, sometimes to nearby points, sometimes to points at infinity (in which case the males moved laterally parallel to the prey). However, although the male hoverfly trajectories are resemblent of motion camouflage, Srinivasan & Davey (1995) were hesitant to conclude that they were the result of motion camouflage rather than an artefact of an alternative tracking behaviour.

More recently Mitzutani et al. (2003) have uncovered evidence that motion camouflage is employed in territorial interactions between conspecific dragonflies. Of 15 trajectories filmed in 3-Dimensions using stereo cameras, 6 showed strong evidence of motion camouflage. Particularly compelling was one trajectory in which the fixed point was apparently located between the shadower and shadowee. This example involved the shadower moving parallel to the shadowee but in the opposite direction (which in light of the other motion camouflage trajectories is difficult to explain otherwise). Later in the same trajectory the shadower appeared to switch to emulating an object at infinity, in a manner similar to that observed in hoverflies (see above).

If the hoverflies and dragonfly flights are indeed instances of motion camouflage behaviour, it is not clear that the technique itself is being used offensively (though in the case of the hoverfly it may be a precursor to an offensive manoeuvre). Rather it appears to be employed to enable the shadower to maintain a position in the locality whilst avoiding eliciting a fleeing (hoverfly) or antagonistic (dragonfly) response in the shadowee, as could be incurred if the shadower's image size or motion passes a threshold level. For example, it would appear that any moving object whose image is about the same size as would be expected of a female hoverfly may be sufficient to elicit a pre-programmed mating attack response in amorous males (Collett & Land, 1978).

Given access to the relative coordinates of shadower, prey and fixed point, perfect motion camouflaged approaches can be calculated mathematically as described by Anderson (2003). Neither hoverfly or dragonfly have access to such explicit positional information. In investigating the level of accuracy that could be achieved with incomplete sensory input information the authors Anderson & McOwan (2003) trained neural network control systems to calculate motion camouflaged approaches in a 3-D environment. To date and to the author's knowledge this has been the only attempt to recreate motion camouflage artificially.

As input,the shadower's neural control model was given the current position of the prey in the shadower's visual field and a short memory of the shadower's recent movements. Thus the shadower was given no explicit indication of the distance to the prey. The movement memory was given with the primary purpose of allowing the shadower to estimate the position of the fixed point (i.e. effectively to allow dead reckoning). Note that insects such as bees and ants are capable of dead-reckoning, using a skylight compass and optic flow based odometer to continually integrate the directions and distances travelled into a representation giving the homebound vector (see Wehner et al., 1996).

The neural control systems were trained to approach pre-determined prey trajectories using a supervised learning algorithm. At each time step during training the parameters of the neural system were adjusted to bring the actual response closer to the target response. Target responses were calculated mathematically with 'gods eye view' knowledge of the exact coordinates of shadower prey and fixed point (see Anderson (2003) for details). The systems were tested using the 2-D digitised flight paths of real hoverflies and 2-D and 3-D artificially generated trajectories as the prey. In each case the control systems were shown to be able to calculate accurate offensive motion camouflaged approaches (see Figure 2). The accuracy of the systems was further tested in the psychophysical experiment mentioned in the following section, demonstrating, in these circumstances, that the effectiveness of the neural control models was not significantly different to that of perfectly camouflaged missiles whose path was calculated mathematically with knowledge of exact coordinates (Anderson & McOwan, 2003b).

  4. HUMANS DECEIVED BY MOTION CAMOUFLAGE

In a psychophysical experiment based on a 3-D computer game, humans became prey, defending themselves against attacks from motion camouflaged missiles (Anderson & McOwan, 2003b). The experiment compared subject 's success in shooting missiles employing different approach strategies as the subject flew along a straight path in a virtual 3-D environment. Alternative missile approach strategies included a homing approach, where the missile constantly moved directly toward the subject and a direct interception approach, where the missiles moved in a straight line to intercept the subject as quickly as possible (See Fig. 3). The experiment's results demonstrated that motion-camouflaged missiles were in general able to get closer to the subject before being shot than the alternative strategies, on average approaching to a distance from the subject ~60% of that achieved by the homing missiles and ~50% that of the direct-interception missiles. The least mean explosion distance per missile type, per player was given by the motion camouflage missiles in 26/30 cases. Screenshots of the game are displayed in Fig. 4).

It is natural to consider methods to develop the strategy of motion camoufage for artificial systems. One possible area of development would be the employment of motion camouflage by a group of individuals. A group of individuals could range from solitary individuals who happen to be shadowing in the same area to a team implementing coordinated complex behaviours. Discussion of a selection of theoretical motion camouflage team manoeuvres and their possible advantages are discussed throughout the remainder of this section. Beforehand, it should be noted that computing camouflaged approaches in real world artificial systems, where shadowers must compensate for mechanical and environmental forces, is a problem far removed from the simulations described above. However, some of the problems faced by hoverflies, dragonflies and Anderson & McOwan's (2003b) neural control systems, operating with incomplete input information may be circumvented by the accuracy of artificial sensory technology. For example, global positioning systems would allow a shadower to accurately estimate its position relative to the fixed point. In circumstances where the prey is not protected by radar jamming apparatus, radar would allow the shadower to accurately estimate the distance of the prey (at risk of exposing the shadower if the radar is active). Furthermore, artificial shadowers would have the opportunity to rapidly communicate detailed information to each other. The ability to communicate would greatly enhance the potential for coordinated team camouflage manoeuvres.

• Application of motion camouflage to real world problems. Likely application areas are in military manoeuvres and computer games.
• As first suggested by Srinivasan & Davey (1995) investigation of possible prey counter strategies to combat camouflaged approaches.

Anderson, A. J. 2003. Sensorimotor neural systems for a predatory stealth behaviour camouflaging motion. PhD Thesis. Queen Mary, University of London.

Anderson, A. J. & McOwan, P. W. 2003. Model of a predatory stealth behaviour camouflaging motion. Proc. R. Soc. Lond. B, 270, 489-495.

Anderson, A. J. & McOwan, P. W. 2003b. Humans deceived by predatory stealth strategy camouflaging motion. Proc. R. Soc. Lond. B. 270. Suppl. 1, S18-S20.

Collett, T. S. & Land, M. F. 1975. Visual control of flight behaviour in the hoverfly, Syritta pipiens L. J. comp. Physiol. 99, 1-66.

Collett, T. S. & Land, M. F. 1978. How hoverflies compute interception courses. J. comp. Physiol. 125, 191-204.

Mizutani, A., Chahl, J. S. & Srinivasan, M. V. 2003. Motion camouflage in dragonflies. Nature, 423, 604.

Srinivasan, M. V. & Davey, M. 1995. Strategies for active camouflage of motion. Proc. R. Soc. Lond. B 259, 19-25.

Wehner, R., Michel, B. & Antonsen, P. 1996. Visual navigation in insects: Coupling of egocentric and geocentric information. J. Exp. Biol. 199, 129-140.

Anderson, A. J. & McOwan, P.W. 2003. Motion camouflage team tactics. to appear at Evolvability & Interaction Symposium, October 2003. [html]

Anderson, A. J. & McOwan, P. W. 2003. Humans deceived by predatory stealth strategy camouflaging motion. Proc. R. Soc. Lond. B, 270 Suppl. 1, s18-S20. [pdf]

Anderson, A. J. & McOwan, P. W. 2003. Model of a predatory stealth behaviour camouflaging motion. Proc. R. Soc. Lond. B 270, 489-495. [pdf]

Anderson, A. J. & McOwan, P. W. 2002. 3D simulation of a sensorimotor stealth strategy for camouflaging motion. International Conference on Neural Information Processing, Singapore, 1805-1810. [pdf]

Anderson, A. J. & McOwan, P. W. 2002. Towards an autonomous motion camouflage control system, International Joint Conference on Neural Networks, Hawaii, 2006-2011. [pdf]