ant search | Stefan M. Popp

These are the first 3 chapters of my PhD thesis. I continue to work on the topic of ant search in 2 other projects.

How do you search for a small item in a huge space you cannot oversee? Previously, the question of how ants search for unknown resources had barely been addressed in the scientific literature. I created to date the largest arena for tracking ants to describe their search patterns and answer questions about the influences of experience and communication.

methods

I let 5 whole colonies of ants search freely in a custom-built large arena which I recorded with 4 cameras to track the ants’ movements. Since I was interested in their search behavior for novel targets, I never added any resources in the arena.

Left: The study species, Temnothorax rugatulus (by Takao Sasaki), Middle: Artificial nest with colony which was placed in the arena underneath a layer of paper. Right: Experimental arena (2x3 m, ~1000x an ant's size).

data processing

I preprocessed the 4 5h long 4K videos with a bash routine, first applying filters in ffmpeg and cutting them up for more accurate tracking (bash script). The processed videos were then fully automatically tracked with TRex (bash script).

For the following data cleaning and annotation, I developed an extensive routine which allowed for semi-automatic aligning and combination of different data sets, outlier exclusion, and tracking error correction. This involved various scripts for the creation of 3D interactive plots with specific functionalities to enable the visual inspection and correction of data. The manual part of the data correction was mainly carried out by undergraduate research assistants which I mentored. The whole data pipeline was thus designed to be user friendly, reproducible, and contained validity checks. I created several tutorials which helped with the training of this complex set of tasks.

Left: Ant paths from one colony evolving over the 5 h of one trial. Right: Tutorial for the joining of erroneously broken up tracks.

results

ants meander regularly

Ants do not merely walk purely randomly, as was previously assumed, but rather meander in regularly alternating smooth turns. This systematic element makes their search patterns more space-filling, and thus efficient, while the randomness conserves flexiblity and robustness to noise and obstacles. (Popp & Dornhaus, 2023)

Left: Track section of a strongly meandering ant. Grid lines are 20 cm apart. Right: Close-up example of a meandering track.

ants explore more near the nest in a novel environment

The study species often relocates their nest, requiring the foragers to get familiar with their new environment for effective navigation and search. Over 3 days after being placed into a new arena, ants moved less only near the nest, where movements are slower and curvier. This reflects a shift from familiarization or marking to searching. However, we don’t see stereotyped ‘learning walks’, like in some desert ant species. (Popp & Dornhaus, 2023)

Left: Heatmaps of all ant locations on the 3 days. Middle Histograms of ant locations by nest distance: on subsequent days, i.e., with increasing familiarity of the ants with the arena, there are fewer exploratory movements near the nest Right: Tracks from the first 5 min on the first day, vs the last 5 min on the last day. There is no measurable difference in movement patterns, i.e. no special 'learning walks'.

ants do not coordinate their search much through chemical signals

Previous work showed that ants walk straighter and faster on surfaces which are marked with chemical footprints of nestmates. We find the opposite effect, which we explain by a) footprints making ants walk straighter, and b) interindividual variation combined with environmental heterogeneity mean reverse causation to the expected. In nature, environmental cues are likely more important than chemical footprints. To be replaced once published: (Popp & Dornhaus, 2023)

Left: Heatmap of all ant points on the 3rd day, with an illustration of interindividual variation in path straightness Right: Heatmap with illustration of the locations of adhesive tape several layers below the topmost paper.

trying to find patterns and recreate tracks using machine learning

As a side project, I collaborated with the Kobourov lab and Alon Efrat to identify and characterize patterns in the movement patterns, like the meanders. We applied dimensionality reduction and clustering,identifying common motifs of the meandering, and explored the creation of tracks via a specialized General Adversarial Network and an ARIMA model

Left: Trajectory created by drawing randomly from the set of identified curve eigenvectors, Middle: Empirical ant track with moving average, which could be used to separate the patterns on different scales Right: Empirical track segment and ARIMA forecast of it.

Code can be found here.

Animal search movements are typically assumed to be mostly random walks, although non-random elements may be widespread. We tracked ants (Temnothorax rugatulus) in a large empty arena, resulting in almost 5 km of trajectories. We tested for meandering by comparing the turn autocorrelations for empirical ant tracks and simulated, realistic Correlated Random Walks. We found that 78% of ants show significant negative autocorrelation around 10 mm (3 body lengths). This means that turns in one direction are likely followed by turns in the opposite direction after this distance. This meandering likely makes the search more efficient, as it allows ants to avoid crossing their own paths while staying close to the nest, avoiding return-travel time. Combining systematic search with stochastic elements may make the strategy less vulnerable to directional inaccuracies. This study is the first to find evidence for efficient search by regular meandering in a freely searching animal.

Central-place foragers must learn the resource and visual landscapes around them when they are in a new environment, to facilitate efficient foraging and navigation. We investigated how the colony-level exploration strategy of Temnothorax rugatulus ants changes over time. For this study, we introduced ants to a novel environment by placing them in a foraging arena and tracking their movements over 3 days for 5 h per day, prohibiting access to the arena at other times. To test whether any changes in movement behavior are due to chemical markings, we replaced the paper floor on the third day. We found that colony-level exploration activity decreased with time, but only within a roughly 1 m radius around the nest, possibly reflecting a shift from familiarization or marking walks to searching. Individuals’ movements overall also became slightly straighter and faster across and within days. However, unlike learning walks of other ant species, T. rugatulus ants did not pause more often when facing toward the nest. Reactions to chemical markings seem to play a minor role in our observed effects, as the exploratory behavior did not reset after the floor cover had been replaced. Thus, ant colony exploration and search behaviors adapt to the familiarity of their environment by becoming more dispersive, possibly aiding in search efficiency. This must be considered in lab studies on the foraging activity and behavior of ants.