Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 134 of the 134 references it cites were found in a registry or an index.

Sophisticated collective foraging with minimalist agents: a swarm robotics test

Open access

Abstract

How groups of cooperative foragers can achieve efficient and robust collective foraging is of interest both to biologists studying social insects and engineers designing swarm robotics systems. Of particular interest are distance-quality trade-offs and swarm-size-dependent foraging strategies. Here, we present a collective foraging system based on virtual pheromones, tested in simulation and in swarms of up to 200 physical robots. Our individual agent controllers are highly simplified, as they are based on binary pheromone sensors. Despite being simple, our individual controllers are able to reproduce classical foraging experiments conducted with more capable real ants that sense pheromone concentration and follow its gradient. One key feature of our controllers is a control parameter which balances the trade-off between distance selectivity and quality selectivity of individual foragers. We construct an optimal foraging theory model that accounts for distance and quality of resources, as well as overcrowding, and predicts a swarm-size-dependent strategy. We test swarms implementing our controllers against our optimality model and find that, for moderate swarm sizes, they can be parameterised to approximate the optimal foraging strategy. This study demonstrates the sufficiency of simple individual agent rules to generate sophisticated collective foraging behaviour.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1007/s11721-019-00176-9

Cites

134 of 134 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 a A picture of a Kilobot with a 3D printed ring [originally designed for the study of Pratissoli et al. (2019)] which considerably improves ARK's performance in terms of tracking and LED colour detection. b Kilobots sense via ARK the presence of virtual pheromone in front of themselves at a distance of ∼ 3 . 5 cm in four 45 ◦ -wide sectors. The virtual sensor indicates the presence or absence of pheromone as binary values, therefore, the Kilobot has no information about the pheromone quantity or concentration difference. In this illustration, pheromone is represented as blue circles, and thus, the virtual sensor readings are [1, 0, 1, 0]. When an exploring Kilobot detects pheromone, it interrupts random exploration and moves towards the detected pheromone. If more than one sector has pheromone (as in the illustration), to decide its motion direction the robot compares the sectors' direction with the depot direction (depot illustrated as a house and direction differences as red and green angles), and moves towards the largest angle (green arrow) (Colour figure online)
Fig. 1 a A picture of a Kilobot with a 3D printed ring [originally designed for the study of Pratissoli et al. (2019)] which considerably improves ARK's performance in terms of tracking and LED colour detection. b Kilobots sense via ARK the presence of virtual pheromone in front of themselves at a distance of ∼ 3 . 5 cm in four 45 ◦ -wide sectors. The virtual sensor indicates the presence or absence of pheromone as binary values, therefore, the Kilobot has no information about the pheromone quantity or concentration difference. In this illustration, pheromone is represented as blue circles, and thus, the virtual sensor readings are [1, 0, 1, 0]. When an exploring Kilobot detects pheromone, it interrupts random exploration and moves towards the detected pheromone. If more than one sector has pheromone (as in the illustration), to decide its motion direction the robot compares the sectors' direction with the depot direction (depot illustrated as a house and direction differences as red and green angles), and moves towards the largest angle (green arrow) (Colour figure online)CC-BY · From the PDF
Fig. 2 Probabilistic finite state machine (PFSM) of the individual robot behaviour. Circles represent states, and arrows are transitions. Robots start exploring the environment through a random walk (RW); when they find a source area they collect an item and return to the depot (GD) laying pheromone according to Eq. (2). Once arrived at the depot, they either turn back (TB) or resume exploration (RW). When explorer robots detect pheromone, they follow it (FP). When robots detect a wall, they avoid it (AO). Controlling individuals through this simple PFSM leads to sophisticated collective foraging dynamics
Fig. 2 Probabilistic finite state machine (PFSM) of the individual robot behaviour. Circles represent states, and arrows are transitions. Robots start exploring the environment through a random walk (RW); when they find a source area they collect an item and return to the depot (GD) laying pheromone according to Eq. (2). Once arrived at the depot, they either turn back (TB) or resume exploration (RW). When explorer robots detect pheromone, they follow it (FP). When robots detect a wall, they avoid it (AO). Controlling individuals through this simple PFSM leads to sophisticated collective foraging dynamicsCC-BY · From the PDF
Fig. 3 Fits of Eq. (7) to data generated by physics-based simulations in order to obtain the model parameters reported in Table 2 in Appendix A. Fitting is performed in case of n = 2 source areas with different quality and equal distance in a , equal quality and different distance in b , and equal quality and distance in c . Data points are represented using symbols, and fits are represented using lines (circles and solid grey lines show collection from source A 1 , while triangles and dash-doted blue lines show collection from source A 2 ). Error bars represent 95% confidence intervals. There is a linear growth for small-to-medium numbers of robots on a path, and a nonlinear decay for medium-to-large numbers of robots on a path. This type of growth-decay curves on population size is widespread in nature (Krause et al. 2002) as in engineering (Gunther 2000) (Colour figure online)
Fig. 3 Fits of Eq. (7) to data generated by physics-based simulations in order to obtain the model parameters reported in Table 2 in Appendix A. Fitting is performed in case of n = 2 source areas with different quality and equal distance in a , equal quality and different distance in b , and equal quality and distance in c . Data points are represented using symbols, and fits are represented using lines (circles and solid grey lines show collection from source A 1 , while triangles and dash-doted blue lines show collection from source A 2 ). Error bars represent 95% confidence intervals. There is a linear growth for small-to-medium numbers of robots on a path, and a nonlinear decay for medium-to-large numbers of robots on a path. This type of growth-decay curves on population size is widespread in nature (Krause et al. 2002) as in engineering (Gunther 2000) (Colour figure online)CC-BY · From the PDF
Fig. 4 Model predictions of yield R depending on worker allocation ρ for equally distant sources d 1 = d 2 = 3 . 5 m, in a , and equally nearby sources d 1 = d 2 = 0 . 6 m, in b , for fixed q 1 = 1andvarying q 2 ∈ { 0 . 5 , 0 . 75 , 1 } . Whensourcesare relatively far ( a ), it is optimal to allocate all workers to the better-quality source area, whereas for source areas in close proximity ( b ) the yield is maximised if the trail between the higher-quality option and depot does not become overcrowded. Parameters: β 1 = β 2 = ¯ β = 0 . 965, T 0 , 1 = T 0 , 2 = ¯ T 0 = 0 . 029, κ 1 = κ 2 = ¯ κ = 2 . 321, and S = 200
Fig. 4 Model predictions of yield R depending on worker allocation ρ for equally distant sources d 1 = d 2 = 3 . 5 m, in a , and equally nearby sources d 1 = d 2 = 0 . 6 m, in b , for fixed q 1 = 1andvarying q 2 ∈ { 0 . 5 , 0 . 75 , 1 } . Whensourcesare relatively far ( a ), it is optimal to allocate all workers to the better-quality source area, whereas for source areas in close proximity ( b ) the yield is maximised if the trail between the higher-quality option and depot does not become overcrowded. Parameters: β 1 = β 2 = ¯ β = 0 . 965, T 0 , 1 = T 0 , 2 = ¯ T 0 = 0 . 029, κ 1 = κ 2 = ¯ κ = 2 . 321, and S = 200CC-BY · From the PDF
Fig. 5 Model predictions of yield R depending on workers allocation ρ for equal qualities q 1 = q 2 = 1. In a , we show the effect of distance on R ; we fixed swarm size S = 200 and A 1 's distance d 1 = 0 . 6 m, and varied of A 2 's distance d 2 ∈ { 0 . 3 m , 0 . 6 m , 0 . 9 m } . Due to the effect of overcrowding, the maximum yield is attained when only a limited number of workers, 10-20%, collected from the nearer source in order to keep the path free from traffic congestion. In b , we report the effect of the swarm size S when it is larger, smaller, or equal to the critical size S c . The sources have equal qualities q 1 = q 2 = 1 and depot-source distances d 1 = d 2 = 0 . 6 m. The critical swarm size S c characterises the effect of overcrowding, i.e. when the swarm is sufficiently large ( S > S c ) it is optimal to keep at least one path with less than 50% workers; otherwise, the effect of overcrowding would decrease the income of resources on both paths. Parameters: β 1 = β 2 = ¯ β = 0 . 965, T 0 , 1 = T 0 , 2 = ¯ T 0 = 0 . 029, κ 1 = κ 2 = ¯ κ = 2 . 321
Fig. 5 Model predictions of yield R depending on workers allocation ρ for equal qualities q 1 = q 2 = 1. In a , we show the effect of distance on R ; we fixed swarm size S = 200 and A 1 's distance d 1 = 0 . 6 m, and varied of A 2 's distance d 2 ∈ { 0 . 3 m , 0 . 6 m , 0 . 9 m } . Due to the effect of overcrowding, the maximum yield is attained when only a limited number of workers, 10-20%, collected from the nearer source in order to keep the path free from traffic congestion. In b , we report the effect of the swarm size S when it is larger, smaller, or equal to the critical size S c . The sources have equal qualities q 1 = q 2 = 1 and depot-source distances d 1 = d 2 = 0 . 6 m. The critical swarm size S c characterises the effect of overcrowding, i.e. when the swarm is sufficiently large ( S > S c ) it is optimal to keep at least one path with less than 50% workers; otherwise, the effect of overcrowding would decrease the income of resources on both paths. Parameters: β 1 = β 2 = ¯ β = 0 . 965, T 0 , 1 = T 0 , 2 = ¯ T 0 = 0 . 029, κ 1 = κ 2 = ¯ κ = 2 . 321CC-BY · From the PDF
Fig. 6 A picture of a 50 real Kilobots experiment with the virtual environment superimposed on the image. The red (bottom-left) source area A 1 has quality Q 1 = 10, while the yellow (top-right) source area A 2 has quality Q 2 = 4. The sources are placed at d 1 = 1 m and d 2 = 0 . 6 m from the central (blue) depot. The (light-blue) shades represent the pheromone trails that the robots deposit and follow. Full videos available as supplementary material (Online Resource 1-9) and at https://www.youtube.com/playlist? list=PLCGKY9OHLZwMaGeB6cxVfxmHwhBFqKF7a (Colour figure online)
Fig. 6 A picture of a 50 real Kilobots experiment with the virtual environment superimposed on the image. The red (bottom-left) source area A 1 has quality Q 1 = 10, while the yellow (top-right) source area A 2 has quality Q 2 = 4. The sources are placed at d 1 = 1 m and d 2 = 0 . 6 m from the central (blue) depot. The (light-blue) shades represent the pheromone trails that the robots deposit and follow. Full videos available as supplementary material (Online Resource 1-9) and at https://www.youtube.com/playlist? list=PLCGKY9OHLZwMaGeB6cxVfxmHwhBFqKF7a (Colour figure online)CC-BY · From the PDF
Fig. 7 A50simulated Kilobot swarm experiment inspired by the ants' double-bridge experiment by Goss et al. (1989) in which two paths, a longer path (1.8 m long) and a shorter path (1.4 m long), connected source to depot. When the simulated swarm had access to only the longer path ( a ), the Kilobots reinforced pheromone on that path and used it for their collections. Instead, when both paths were available ( b ), the Kilobots disregarded the longer path and (almost exclusively) used the shorter for their collections. c Shows the number of robots on the two paths at the end of one simulated hour (boxes range from 1st to 3rd quartile of the data from 100 simulations and indicate the median with a horizontal line, the whiskers extends to 1.5 IQR). The individual Kilobots cannot follow a pheromone gradient nor detect any difference in pheromone concentration. Despite their limited individual capabilities, the robot swarm shows (in certain experimental conditions) behaviour similar to ants' colonies, which instead rely on much higher cognitive abilities at the individual level
Fig. 7 A50simulated Kilobot swarm experiment inspired by the ants' double-bridge experiment by Goss et al. (1989) in which two paths, a longer path (1.8 m long) and a shorter path (1.4 m long), connected source to depot. When the simulated swarm had access to only the longer path ( a ), the Kilobots reinforced pheromone on that path and used it for their collections. Instead, when both paths were available ( b ), the Kilobots disregarded the longer path and (almost exclusively) used the shorter for their collections. c Shows the number of robots on the two paths at the end of one simulated hour (boxes range from 1st to 3rd quartile of the data from 100 simulations and indicate the median with a horizontal line, the whiskers extends to 1.5 IQR). The individual Kilobots cannot follow a pheromone gradient nor detect any difference in pheromone concentration. Despite their limited individual capabilities, the robot swarm shows (in certain experimental conditions) behaviour similar to ants' colonies, which instead rely on much higher cognitive abilities at the individual levelCC-BY · From the PDF
Fig. 8 Simulation results showing the adaptivity of the system. We measured the number of collected items in a and the number of robots on each path in b for the two source areas, the superior A 1 and inferior A 2 , both at equal distance d 1 = d 2 = 1 m. We kept the same quality ratio, i.e. Q 2 / Q 1 = 0 . 4, but varied the absolute value of the objects (indicated on the x axis). All experiments were conducted with swarms of S = 50 Kilobots and an intermediate value of α = 0 . 85 in Eqs. (2) and (3). Boxes range from 1st to 3rd quartile of the data from 100 simulations and indicate the median with a horizontal line; the whiskers extend to 1.5 IQR. Having a constant range (dark boxplots) shows good results only if the predefined range matches the actual range of the environment (central experiment). Instead, an adaptive strategy allows the swarm to exploit resources as a function of the their relative qualities in a range adapted to the environment
Fig. 8 Simulation results showing the adaptivity of the system. We measured the number of collected items in a and the number of robots on each path in b for the two source areas, the superior A 1 and inferior A 2 , both at equal distance d 1 = d 2 = 1 m. We kept the same quality ratio, i.e. Q 2 / Q 1 = 0 . 4, but varied the absolute value of the objects (indicated on the x axis). All experiments were conducted with swarms of S = 50 Kilobots and an intermediate value of α = 0 . 85 in Eqs. (2) and (3). Boxes range from 1st to 3rd quartile of the data from 100 simulations and indicate the median with a horizontal line; the whiskers extend to 1.5 IQR. Having a constant range (dark boxplots) shows good results only if the predefined range matches the actual range of the environment (central experiment). Instead, an adaptive strategy allows the swarm to exploit resources as a function of the their relative qualities in a range adapted to the environmentCC-BY · From the PDF
Fig. 9 Effect of the modulation of the parameter α from Eqs. (2) and (3) to favour nearer source areas ( α = 0), to favour the best-quality sources ( α = 10), or to balance the distance-quality trade-off (0 < α < 10). Results of α = 0 are shown in light-grey, α = 0 . 85 in dark-grey, and α = 10 in black. We report the results for simulations and physical robots experiments of one hour each in scenarios with n = 2 sources. We excluded the initial exploration phase and indicate mean values for the last 30 min. Physical robots results are indicated as solid symbols with vertical bars indicating the 95% confidence intervals of 3 runs for each condition (the symbols are slightly shifted to avoid bar overlaps but all represent results for d 2 = 0 . 6 m). Lines represent the mean of 100 simulations (shaded areas are 95% confidence intervals). Source A 1 had quality Q 1 = 10 and was located at distance d 1 = 1 m; source A 2 had quality Q 2 = 4 and varying distance d 2 ∈ [ 0 . 5 , 1 . 0 ] m. We report the rate of collected items per minute in a , the mean number of robots on each path in b , and the rate per minute of collected items weighted by the normalised quality q 1 = 1 . 0 and q 2 = 0 . 5 in c . Individual robots can locally modulate the decentralised parameter α to lead the swarm to a range of different collective responses, e.g. selecting almost exclusively the best-quality source (high α ) or balancing the distance-quality trade-off (low α ). Physical robots are less efficient than simulations; however, ordering between sources is preserved; this confirms the effects of α -modulation observed in simulation
Fig. 9 Effect of the modulation of the parameter α from Eqs. (2) and (3) to favour nearer source areas ( α = 0), to favour the best-quality sources ( α = 10), or to balance the distance-quality trade-off (0 < α < 10). Results of α = 0 are shown in light-grey, α = 0 . 85 in dark-grey, and α = 10 in black. We report the results for simulations and physical robots experiments of one hour each in scenarios with n = 2 sources. We excluded the initial exploration phase and indicate mean values for the last 30 min. Physical robots results are indicated as solid symbols with vertical bars indicating the 95% confidence intervals of 3 runs for each condition (the symbols are slightly shifted to avoid bar overlaps but all represent results for d 2 = 0 . 6 m). Lines represent the mean of 100 simulations (shaded areas are 95% confidence intervals). Source A 1 had quality Q 1 = 10 and was located at distance d 1 = 1 m; source A 2 had quality Q 2 = 4 and varying distance d 2 ∈ [ 0 . 5 , 1 . 0 ] m. We report the rate of collected items per minute in a , the mean number of robots on each path in b , and the rate per minute of collected items weighted by the normalised quality q 1 = 1 . 0 and q 2 = 0 . 5 in c . Individual robots can locally modulate the decentralised parameter α to lead the swarm to a range of different collective responses, e.g. selecting almost exclusively the best-quality source (high α ) or balancing the distance-quality trade-off (low α ). Physical robots are less efficient than simulations; however, ordering between sources is preserved; this confirms the effects of α -modulation observed in simulationCC-BY · From the PDF
Fig. 10 Comparison of model with simulations and experiments: Total yield R as a function of the normalised swarm allocation ρ/ρ w and the number of worker robots ρ w S . We report the predicted yield R from the model of Eq. (4) as a colour heatmap, and we overlay robot simulations for three strategies: distance-selective α = 0 (circles), distance-quality trade-off α = 0 . 85 (diamonds), and quality-selective α = 10 (triangles). We report simulations for swarm sizes S = 50 (cyan), S = 100 (green), S = 200 (purple) and S = 500 (white). Under the model's assumptions, the simulated robot swarm performs best for S = 200 and α = 0 . 85 ( R = 150 . 6 m -2 ) in a , α = 10 ( R = 177 . 1 m -2 ) in b and α = 10 ( R = 120 . 4 m -2 ) in c . Swarms of large size ( S = 500) do not achieve good performance as they equally exploit both sources and do not avoid overcrowding. The star symbol in c was obtained from three experiments with 200 Kilobots assuming α = 0 . 85 (see online videos). Error bars represent 95% confidence intervals. Parameters: β j , T 0 , j and κ j are given in Table 2 (Colour figure online)
Fig. 10 Comparison of model with simulations and experiments: Total yield R as a function of the normalised swarm allocation ρ/ρ w and the number of worker robots ρ w S . We report the predicted yield R from the model of Eq. (4) as a colour heatmap, and we overlay robot simulations for three strategies: distance-selective α = 0 (circles), distance-quality trade-off α = 0 . 85 (diamonds), and quality-selective α = 10 (triangles). We report simulations for swarm sizes S = 50 (cyan), S = 100 (green), S = 200 (purple) and S = 500 (white). Under the model's assumptions, the simulated robot swarm performs best for S = 200 and α = 0 . 85 ( R = 150 . 6 m -2 ) in a , α = 10 ( R = 177 . 1 m -2 ) in b and α = 10 ( R = 120 . 4 m -2 ) in c . Swarms of large size ( S = 500) do not achieve good performance as they equally exploit both sources and do not avoid overcrowding. The star symbol in c was obtained from three experiments with 200 Kilobots assuming α = 0 . 85 (see online videos). Error bars represent 95% confidence intervals. Parameters: β j , T 0 , j and κ j are given in Table 2 (Colour figure online)CC-BY · From the PDF
Fig. 11 Total yield R as a function of the normalised swarm allocation ρ/ρ w and the number of worker robots ρ w S in the collision-free condition. We removed the effect of physical interactions (i.e. collisions between robots) that may cause traffic congestions, and we report the predicted yield R from model (4) as a colour heatmap, and we overlay robot simulations for three strategies: distance-selective α = 0 (circle), distancequality trade-off α = 0 . 85 (diamond), and quality-selective α = 10 (triangle). We report simulations for swarm sizes S = 50 (cyan), S = 100 (green), S = 200 (purple) and S = 500 (white). Without collision, the predicted best strategy is allocation of all workers to the best-quality or closest source area. The collisionfree simulations approximate such result when the corresponding strategy is activated, e.g. quality-selective α = 10 (triangle) in panel (a) and the distance-selective α = 0 (circle) in panel (b). Error bars represent 95% confidence intervals. Parameters: β j , T 0 , j and κ j are given in Table 2 (Colour figure online)
Fig. 11 Total yield R as a function of the normalised swarm allocation ρ/ρ w and the number of worker robots ρ w S in the collision-free condition. We removed the effect of physical interactions (i.e. collisions between robots) that may cause traffic congestions, and we report the predicted yield R from model (4) as a colour heatmap, and we overlay robot simulations for three strategies: distance-selective α = 0 (circle), distancequality trade-off α = 0 . 85 (diamond), and quality-selective α = 10 (triangle). We report simulations for swarm sizes S = 50 (cyan), S = 100 (green), S = 200 (purple) and S = 500 (white). Without collision, the predicted best strategy is allocation of all workers to the best-quality or closest source area. The collisionfree simulations approximate such result when the corresponding strategy is activated, e.g. quality-selective α = 10 (triangle) in panel (a) and the distance-selective α = 0 (circle) in panel (b). Error bars represent 95% confidence intervals. Parameters: β j , T 0 , j and κ j are given in Table 2 (Colour figure online)CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 133 references

Research works printed but not linked: 1

  1. Orians, G. H., & Pearson, N. E. (1979). On the theory of central place foraging. In D. J. Horn, R. D. Mitchell, & G. R. Stairs (Eds.), Analysis of ecological systems (pp. 154–177). Columbus: The Ohio State University Press.

Cited by

See all 70

Metadata

Authors
Talamali, M. S., Bose, T., Haire, M., Xu, X., Marshall, J. A. R., & Reina, A.
Affiliations
  • University of Sheffield (GB)
  • Sheffield Hallam University (GB)
Corresponding author
Talamali, Mohamed S.
Venue
Swarm Intelligence
Published
2019-10-10
Volume
14
Issue
1
Pages
25-56
ISSN
  • 1935-3812 (ISSN-L)
  • 1935-3820
OpenAlex
W2979406491
Cited by
70
References
133
Citations by year
2026: 12 · 2025: 6 · 2024: 10 · 2023: 10 · 2022: 9 · 2021: 14 · 2020: 7 · 2019: 2
FWCI
4.50
Type
journal article
Language
en
Open access
Licence
Locations
  • Swarm Intelligence (publishedVersion, cc-by)
  • White Rose Research Online (University of Leeds, The University of Sheffield, University of York) (acceptedVersion)
  • SHURA (Sheffield Hallam University Research Archive) (Sheffield Hallam University) (acceptedVersion)
  • Dépôt institutionnel de l'Université libre de Bruxelles (Université Libre de Bruxelles) (submittedVersion, other-oa)
Funders
H2020 European Research Council (647704)
Topics
  • Physical Sciences › Engineering › Mechanical Engineering › Modular Robots and Swarm Intelligence
  • Physical Sciences › Computer Science › Artificial Intelligence › Metaheuristic Optimization Algorithms Research
  • Physical Sciences › Engineering › Control and Systems Engineering › Distributed Control Multi-Agent Systems
Keywords
group foraging, swarm robotics, virtual pheromones, swarm size, optimal foraging theory, ant foraging behavior, resource quality, overcrowding

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Sophisticated collective foraging with minimalist agents: a swarm robotics test. https://app.cochat.ai/papers/10.1007/s11721-019-00176-9
The paper
Talamali, M. S., Bose, T., Haire, M., Xu, X., Marshall, J. A. R., & Reina, A. (2020). Sophisticated collective foraging with minimalist agents: a swarm robotics test. Swarm Intelligence. https://doi.org/10.1007/s11721-019-00176-9

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.