Signal design facilitates recognition but not detection: A field test using robots

Previously posted on Anole Annals:

The use of programmable robots (‘mechanical models’ is more accurate) to minimise disturbance while observing wildlife, or to run behavioural experiments in the field, has slowly increased in the last decade and studies across many taxa have utilized this approach (Martins et al., 2005; Partan et al., 2009; Cianca et al., 2013; Macedonia et al., 2013; Clark et al., 2015). I’d argue that “robots” are one for the most important tools for behavioural ecologists studying communication or display behaviour, as they are one of the few ways in which we can conduct field-based experiments – mimicking or manipulating animal behaviour, colour or morphology in any way – in the animal’s natural environment.

We recently published a paper in the Journal of Evolutionary Biology, using robots in playback experiments to test the importance of ornament design for signal detection and conspecific recognition.

Many factors potentially affect signal design, including the need for rapid signal detection and the ability to identify the signal as conspecific. As understanding these different sources of selection on signal design is essential in the larger goal of explaining the evolution of both signal complexity and signal diversity, here we assessed the relative importance of detection and recognition for signal design in the Black-bearded gliding lizard, Draco melanopogon (fig. 1). Lizards of the species-rich genus Draco use large extendible dewlaps for communication, that differ in colour pattern and size between species – in a similar fashion to the anoles.

AA_figure 1


Figure 1 A. Male D. melanopogan, dewlap naturally extended (image a still from behavioural trials) and the angle of dewlap extension as measured from still; B. robot, dewlap treatments (Bi) solid colour and Bii) two-coloured); and C. artificially extended dewlaps of a male and female D. melanopogan.

To test whether the dewlap colour and pattern function more to facilitate 1. signal detection and 2. conspecific recognition, we presented free-living lizards with robots displaying dewlaps of six different designs, varying in the proportion of the black and white components.

In this case, our robots were just ‘visual flags’ that mimicked the dewlap size and shape, as well as the speed and display pattern of live Draco melanopogan lizards (video 1). Having only the dewlap / visual flag and not the rest of the lizard body allowed us to look solely at the salience of the dewlap colour and pattern itself – without adding any identifying or qualifying information in the form of a body.

Video 1: ‘The floating dewlap’

Our experiment had six colour treatments ranging from “natural” (population typical design, fig. 1) to unnatural (wrong colour, no pattern) – and from very conspicuous (high internal contrast and high contrast against the background for each colour) to very inconspicuous (matching the luminance of the background). Thus, we could test both the ‘detection’ and ‘conspecific recognition’ hypotheses with the same set of treatments.

Predictions for Hypothesis 1: We predicted that should the dewlap colour pattern function in signal detection, that more conspicuous dewlap treatments would be detected sooner than less conspicuous dewlaps. Each of the two-coloured treatments were more conspicuous than the single-coloured treatments, as they had the same high contrast black and white elements, but they also had the high internal contrast of the black against the white (75.02 JND). Provided the receiver has sufficient visual acuity at the viewing distance to be able to distinguish the two colours from one another, internal contrast increases signal conspicuousness, and the more equal the two adjacent colour patches are in size (i.e. 50% of the dewlap black – 50% of the dewlap white) the greater the internal contrast. There is no existing data on the visual acuity of Draco lizards, so for this experiment we stuck to the natural dewlap size and viewing distances, with small oscillations around the natural proportions of black and white.

Therefore the 50-50 treatment was predicted to be detected most rapidly, followed by the 70-30 and the 90-10 treatment. The single-coloured dewlaps (black, white and grey) lacked internal contrast and so we predicted they would all be detected with a greater latency than the two-coloured treatments. The grey treatment was the least conspicuous so we expected that the be the least readily detected (table 1).


Table 1. Achromatic contrast values (JND – two-coloured treatments are taken as a weighted average, which assumes the two colours can be resolved); predictions for the detection hypothesis; and predictions for the conspecific recognition hypothesis.

Predictions for Hypothesis 2: Should dewlap colour pattern function in conspecific recognition, we predicted that the most natural looking dewlap (those closest in colour and pattern to the population typical design) would elicit the greatest response from live lizards. Therefore, we expected the 70-30 dewlap to elicit the greatest response, followed by the 50-50 and the 90-10 dewlap treatments. Then come the single-colour treatments. The black and the white had at least one correct colour, but lacked that ‘pattern’, whilst the grey lacked both colour and pattern. So we predicted a lesser response for the solid black treatment and the solid white treatments, compared to the two-coloured treatments, and we expected the solid grey to elicit the least response from conspecifics (table 1).

In planning this experiment, we first visited the sight and collected behavioural, colour and habitat data for the population which we used to build our robots. The robots were programmed with the population-typical display pattern and speed, and the treatment colours were a close match to natural black and white of dewlap. We then took our robots back into the field. When we came across free-living D. melanopogan lizards we set the robots up on long poles a-top tripods (giving them over a 3m perch height) in the periphery of the lizard, at the standard neighbour distance of 4-5 meters (fig. 2). The robots had a delayed-start ‘on’ button, giving us time to set-up the video camera on another tripod and then let everything settle for 5-6 minutes before the robot started displaying. The footage was then analysed for latency to orientate (eye or head movements in the direction of the displaying robot) and response intensity (number of dewlap extensions after detecting the robot).


Figure 2. Robot and observer set-up in the field.

Results 1. Figure 3 shows the latency for lizards to orientate to the 6 different treatments and whilst a treatment*robot distance model was my best supported – you can see from the graph that there are no real differences between the treatments in how quickly they’re being detected. These results suggest that initial signal detection (simulation of the visual grasp reflex) is the same regardless of signal contrast. This was a surprise to me as I had always assumed high contrast colours were important for signal detection – instead I guess dewlap movement is the most important aspect for signal detection, at least in a natural setting.


Figure 3. Latency to orientate to each treatment as a function of robot distance from the focal lizard (males and females combined), A) single-coloured treatments and B) two-coloured treatments. Each dot represents the point at which an individual lizard detected the robot. Trend lines represent the mean orientation time as a function of robot distance, with grey bands representing 95 % confidence intervals.

Results 2. In terms of conspecific recognition, as expected, we found males responded with the greatest intensity to the 70-30 ‘natural’ dewlap pattern (figure 4). Response intensity for females, however, was not consistent with males. I suggest this is not because they don’t recognize that 70-30 dewlap as conspecific, but because female response intensity isn’t as clear cut as the males. Males are territorial and constantly monitor their territories for intruders. Females do the same for other females, but the ‘natural’ dewlap treatment was really the ‘natural male’ treatment, and female response to a male in their territory might depend on several factors, such as mating receptivity. Also the lizard body was missing, and female lizards can be picky about stuff like that in their mates. We also found that males were more likely to respond to any treatment when they had a neighbour in close proximity – so even their responses are dependent on the social context.


Figure 4. Boxplot showing the number of dewlap displays performed by lizards in 2.5 minutes following detection for (A) males and (B) females.

Unusually, the second greatest male response was to the grey treatment – which I had pegged as the least ‘natural’. The most likely explanation for this is that lizards may not be able to resolve the two colours of the D. melanopogon dewlap in natural conditions, for distances at which broadcast displays are typically given. An artifact of sticking to the achromatic palette for the unnatural, inconspicuous treatment (grey) was that it ended up closely resembling the two colours of the natural dewlap should they appear as a ‘blended average’ (i.e. if the lizard cannot resolve the black and white patches from one another). Perhaps the two colours are more important in close-range interactions, such as aggressive competition and courtship – in which case it’s not really surprising that the movement of the dewlap is more important than colour contrast for signal detection!

In summary, we found no effect of dewlap brightness or design on the time it took for a lizard to detect the robot, consistent with the view that initial detection is likely to be primarily elicited by movement rather than specific colour or pattern. However, males responded with a greater intensity to the dewlap treatment that most resembled the natural dewlap colour and design of the species, suggesting the colour and pattern of the dewlap play a strong role in conspecific recognition.

Cianca, V., Bartolini, T., Porfiri, M., Macrì, S. & Scharlemann, J. 2013. A Robotics-Based Behavioral Paradigm to Measure Anxiety-Related Responses in Zebrafish. PLoS One 8: e69661. Public Library of Science.
Clark, D.L., Macedonia, J.M., Rowe, J.W., Stuart, M.A., Kemp, D.J. & Ord, T.J. 2015. Evolution of displays in Galapagos lava lizards: comparative analyses of signallers and robot playbacks to receivers. Anim. Behav. 109: 33–44.
Macedonia, J.M., Clark, D.L., Riley, R.G. & Kemp, D.J. 2013. Species recognition of color and motion signals in Anolis grahami: Evidence from responses to lizard robots. Behav. Ecol. 24.
Martins, E.P., Ord, T.J. & Davenport, S.W. 2005. Combining motions into complex displays: playbacks with a robotic lizard. Behav. Ecol. Sociobiol. 58: 351–360.
Partan, S.R., Larco, C.P. & Owens, M.J. 2009. Wild tree squirrels respond with multisensory enhancement to conspecific robot alarm behaviour. Anim. Behav. 77: 1127–1135.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s