272
L. Musiolek et al.
2.2 Procedure of the Behavioural Experiment
Our sample consisted of 22 adult female Trinidadian guppies (Poecilia reticulata), marked individually with colour tags for better identification [1]. The
Robofish trajectories were programmed open-loop as shown in Fig. 1a. After two
minutes of acclimatization time inside the cylinder at one end of the test tank,
its door was opened gently. As soon as the fish emerged by one body length,
we started the Robofish on the trajectory to its goal. All trajectories (and thus
all trials) took around 25 s to complete. All fish were tested on five consecutive
days, making our basic design a mixed “two by five” one, with two treatments
as between-subjects factor and five within-subject repetitions. The experiment
complied with the current German law approved by LaGeSo (G0117/16 granted
to D. Bierbach).
2.3 Analysis
All trials were video recorded with a Basler Ace 2K USB3 camera (50fps) and
subsequently tracked with Ethovision XT 14 (same sample fps). From the tracks,
we extracted the following response variables in order to describe social interactions among live fish and replicas: (I) Inter-individual distance (IID), the average
distance between a fish and Robofish over all frames of a trial. (II) Normalized
cross-correlations (Xcorr) between the frame-wise displacement vectors of the
two subjects. Xcorr were computed for time lags of up to 10 s in either direction
between the fish and robot displacement time series and the highest of these
values was chosen. We fitted a linear mixed-effects model including treatment
and experiment day as categorical fixed effects, their interaction, as well as random intercepts for the individual fish for both of our dependent variables [2].
Complete data and analyses are available upon request.
3 Results
A significant interaction between treatment and day for IID (χ
2 (4, N = 22) =
12.11, p = 0.017, Fig. 2a) prompted us to perform the same analysis also on a
data subset containing only the first three days. Here, we found a significant
main effect for treatment (χ
2 (1, N = 22) = 8.38, p = 0.004), with fish in the
social treatment being on average 8.8 cm closer to the replica than those in the
cuboid (nonsocial) treatment. Day had a significant effect on IID (χ
2 (4, N =
22) = 17.05, p = 0.002) across all five days, but not across only the first three
days. While no significant effects were detectable for Xcorr across all five days,
on the first three days fish tested with lifelike replicas tended to have higher
Xcorr compared to those tested with cuboid replicas (β = 0.09, SE = 0.09, χ
2 (1,
N = 22) = 2.88, p = 0.09, see Fig. 2b).
L. Musiolek et al.
2.2 Procedure of the Behavioural Experiment
Our sample consisted of 22 adult female Trinidadian guppies (Poecilia reticulata), marked individually with colour tags for better identification [1]. The
Robofish trajectories were programmed open-loop as shown in Fig. 1a. After two
minutes of acclimatization time inside the cylinder at one end of the test tank,
its door was opened gently. As soon as the fish emerged by one body length,
we started the Robofish on the trajectory to its goal. All trajectories (and thus
all trials) took around 25 s to complete. All fish were tested on five consecutive
days, making our basic design a mixed “two by five” one, with two treatments
as between-subjects factor and five within-subject repetitions. The experiment
complied with the current German law approved by LaGeSo (G0117/16 granted
to D. Bierbach).
2.3 Analysis
All trials were video recorded with a Basler Ace 2K USB3 camera (50fps) and
subsequently tracked with Ethovision XT 14 (same sample fps). From the tracks,
we extracted the following response variables in order to describe social interactions among live fish and replicas: (I) Inter-individual distance (IID), the average
distance between a fish and Robofish over all frames of a trial. (II) Normalized
cross-correlations (Xcorr) between the frame-wise displacement vectors of the
two subjects. Xcorr were computed for time lags of up to 10 s in either direction
between the fish and robot displacement time series and the highest of these
values was chosen. We fitted a linear mixed-effects model including treatment
and experiment day as categorical fixed effects, their interaction, as well as random intercepts for the individual fish for both of our dependent variables [2].
Complete data and analyses are available upon request.
3 Results
A significant interaction between treatment and day for IID (χ
2 (4, N = 22) =
12.11, p = 0.017, Fig. 2a) prompted us to perform the same analysis also on a
data subset containing only the first three days. Here, we found a significant
main effect for treatment (χ
2 (1, N = 22) = 8.38, p = 0.004), with fish in the
social treatment being on average 8.8 cm closer to the replica than those in the
cuboid (nonsocial) treatment. Day had a significant effect on IID (χ
2 (4, N =
22) = 17.05, p = 0.002) across all five days, but not across only the first three
days. While no significant effects were detectable for Xcorr across all five days,
on the first three days fish tested with lifelike replicas tended to have higher
Xcorr compared to those tested with cuboid replicas (β = 0.09, SE = 0.09, χ
2 (1,
N = 22) = 2.88, p = 0.09, see Fig. 2b).