Data and code for: Spontaneous problem-solving in bumblebees
Data files
Apr 17, 2026 version files 17.63 MB
-
BinaryTest_Ins8.csv
514 B
-
Ins8_Exp3.csv
1.57 KB
-
Insight.csv
1.91 KB
-
PercFB.csv
737 B
-
PFB.csv
1.10 KB
-
README.md
9.76 KB
-
SPONTANEOUS_PROBLEM-SOLVING_IN_BUMBLE_BEES.R
28.28 KB
-
successful.zip
12.60 MB
-
unsuccessful.zip
4.98 MB
Abstract
Problem-solving using novel solutions without explicit training is often considered a hallmark of cognitive flexibility. We investigated whether bumble bees (Bombus terrestris) could solve a novel object manipulation task spontaneously. Bees trained to associate a blue ring ('flower'), on the floor with a reward successfully moved a ball underneath a flower relocated to the ceiling, in order to reach the flower. In control experiments, where the flower was out of sight when ball movement began and remained hidden during transport, bees still succeeded in the task. These results suggest that these were goal-directed actions rather than reinforcement-based associations driven by perceptual feedback. Our findings provide evidence that bumble bees can exhibit spontaneous problem-solving, challenging the notion that such advanced cognitive abilities are exclusive to large-brained vertebrates.
Dataset DOI: 10.5061/dryad.w3r22816v
Description of the data and file structure
These data were collected to investigate whether bumble bees (Bombus terrestris) can solve a novel object-manipulation task in a goal-directed manner without direct training on the solution. Across three experiments, individual bees were tested in controlled laboratory arenas where they had to move a ball to access a previously learned reward location (“flower”).
The experimental design manipulated prior experience (pre-training) and the availability of perceptual feedback using physical barriers and visually occluded goal locations. Behavioral responses were recorded at the individual level, including task success, interaction with the object, and movement trajectories.
The dataset enables analysis of how prior information and sensory constraints influence problem-solving performance and whether behavior reflects spontaneous, goal-directed action rather than trial-and-error or perceptual guidance.
Files and variables
File: Ins8_Exp3.csv
Description:
This file contains individual-level behavioral data from Experiment 3, in which bees performed a binary choice task under visual occlusion. Bees were required to move a ball into one of two compartments, only one of which corresponded to the previously learned (but visually hidden) flower location. The dataset includes measures of task success, inspection behavior, and ball manipulation.
Variables:
- BeeID_orig: Original identifier assigned during data collection
- BeeID: Processed/standardized individual identifier used in analyses
- Colony: Colony identity of the bee
- Success: Binary outcome (1 = ball moved into correct compartment, 0 = incorrect)
- Side: Side of the arena where the flower was located (left/right)
- LastcheckFlowerSide: Whether the last inspection before ball movement was directed to the flower compartment (1 = yes, 0 = no)
- NumberOfChecksFlowerCompArea: Number of inspections directed toward the flower compartment area
- NumberOfChecksEmptyCompArea: Number of inspections directed toward the empty compartment area
- PropCheckFlowerCompArea: Proportion of inspections directed toward the flower compartment (NumberOfChecksFlowerCompArea / NumberOfChecks)
- NumberOfChecks: Total number of inspections
- TimeSpentWithTheBall: Total time (s) spent interacting with the ball
- StraightforwardMove: Whether the ball trajectory was direct (1 = yes, 0 = no)
- CorrectionMoves: Whether the bee made corrective adjustments during ball movement (1 = yes, 0 = no)
- CheckBetweenMoving: Whether inspections occurred between ball movements (1 = yes, 0 = no)
- Major corrections: Whether the bee performed major correction sequences (1 = yes, 0 = no)
File: SPONTANEOUS_PROBLEM-SOLVING_IN_BUMBLE_BEES.R
Description:
R script used for data processing, statistical analyses, and figure generation. Includes generalized linear mixed models (GLMMs), model diagnostics, and visualization code corresponding to all experiments.
File: BinaryTest_Ins8.csv
Description:
Dataset from Experiment 3 summarizing binary choice performance. Each row represents one individual bee tested for its ability to move the ball into the correct (flower-associated) compartment under visual occlusion.
Variables:
- BeeID_orig: Original identifier assigned during data collection
- BeeID: Processed/standardized individual identifier
- Colony: Colony identity
- Success: Binary outcome (1 = correct compartment, 0 = incorrect)
- Side: Side of the arena where the flower was located (left/right)
File: PFB.csv
Description:
Dataset from Experiment 2b (three-barrier task), where bees were required to move the ball through multiple barriers. Includes both Treatment (flower present) and Control (no flower) groups.
Variables:
- BeeID_R: Original/raw identifier
- BeeID: Processed individual identifier
- Colony: Colony identity
- Treatment: Experimental condition (Treatment = flower present, Control = no flower)
- Success: Binary outcome (1 = ball moved to final compartment, 0 = not)
File: PercFB.csv
Description:
Dataset from Experiment 2a (single-barrier task), including detailed movement and directional performance measures during ball transport relative to the flower location.
Variables:
- ID: Individual identifier
- Colony: Colony identity
- Success: Binary outcome (1 = successful task completion, 0 = not)
- Side: Side of the flower relative to the arena
- MoveCorrect: Number of ball movements toward the correct (flower) side
- MoveUncorrect: Number of movements toward the incorrect side
- Prop: Proportion of correct movements relative to total movements
- MovesTotal: Total number of ball movements
- Prop1: Alternative proportional measure of directional movement (as defined in analysis script)
- Order: Trial or processing order
File: Insight.csv
Description:
Dataset from Experiment 1, containing behavioral performance measures across different pre-training treatments. Used to assess spontaneous problem-solving in a novel object-manipulation task.
Variables:
- BeeID: Individual identifier
- Colony: Colony identity
- Treatment: Pre-training condition (All Information, Restricted Information, True Control)
- Success: Binary outcome (1 = task solved, 0 = not)
- SolvingTime: Time (s) to complete the task
- RollingTime: Time (s) spent interacting with the ball
- FinalTime: Total trial duration (s)
- Attempts: Number of attempts to reach the flower
- Moves: Number of discrete ball movements
- Position: Final or intermediate ball position category
- Order: Trial or scoring order
- Arena: Arena identifier (if multiple setups used)
- Treatment Key: Encoded representation of treatment group
Files: successful.zip and unsuccessful.zip (tracking data)
Description:
These compressed archives contain time-resolved positional tracking data from Experiment 3. Each CSV file corresponds to a single trial.
- successful.zip: trials where bees solved the task
- unsuccessful.zip: trials where bees failed
These data enable analysis of movement trajectories, interaction dynamics, and behavioral patterns.
File structure (within zip archives)
Each CSV file contains frame-by-frame tracking data:
- Frame: Frame number
- Time: Time elapsed (seconds, if available)
Tracking variables are organized using DeepLabCut’s multi-level header:
Header levels:
- scorer: Name of trained DeepLabCut model
- bodyparts: Tracked anatomical/object points
- coords: Data type (x, y, likelihood)
Tracked points
Bee:
- Head: Head position
- tail: Tip of abdomen
Used to estimate body orientation and movement direction.
Ball:
- ball_top, ball_bottom, ball_left, ball_right, ball_center
Multiple points improve robustness under occlusion and allow reconstruction of ball movement and position.
Coordinate variables
For each tracked point:
- x: Horizontal position (pixels)
- y: Vertical position (pixels)
- likelihood: Confidence score (0–1)
Notes on likelihood values
- High values (~1): reliable detection
- Low values (~0): uncertain detection
- Users may filter data based on likelihood thresholds
Coordinate system
- 2D coordinates extracted from video recordings
- Units correspond to pixels unless calibrated
- Consistent within each trial
Linking tracking data to behavioral data
Tracking files correspond to individuals in:
- Ins8_Exp3.csv
- BinaryTest_Ins8.csv
Matching variables:
- BeeID
- Success (determines zip folder)
Potential derived measures
Users can compute:
- Movement trajectories
- Distance between the bee and the ball
- Directionality relative to the goal
- Latency to interaction
- Path efficiency and straightness
- Behavioral transitions (inspection → manipulation)
Code/software
All data files can be viewed using standard spreadsheet software (e.g., Microsoft Excel, LibreOffice Calc) or imported into R for analysis.
Statistical analyses and data processing were conducted in R (RStudio, version 2026.01.0). The following R packages were used: glmmTMB, DHARMa, broom.mixed, pROC, ggplot2, dplyr, tidyr, ggeffects, tidyverse, MASS, viridis, MuMIn, performance, and emmeans.
Behavioral tracking data (CSV files containing positional coordinates) can be visualized and analyzed in R or other software capable of handling time-series coordinate data. Tracking data were originally generated using AnimalTA (version 3.2.2) and DeepLabCut (version 3.0.0rc9).
The file SPONTANEOUS_PROBLEM-SOLVING_IN_BUMBLE_BEES.R contains all scripts used for data cleaning, statistical modeling (generalized linear mixed models), and figure generation. The workflow consists of importing CSV data files, preprocessing variables, fitting statistical models, and producing visualizations corresponding to the main analyses reported in the manuscript.
Access information
All data supporting the findings of this study are provided within this Dryad repository. The data were generated as part of the present study and are not derived from external datasets. No restrictions apply to the use of these data beyond the terms of the Dryad Digital Repository license.
Data collection and processing
Study system and housing
Experiments were conducted on Bombus terrestris colonies housed in bipartite wooden nest boxes (40 × 28 × 11 cm) connected via a transparent acrylic tunnel to a flight arena. Colonies were maintained under controlled laboratory conditions (19–22 °C; 12 h light–dark cycle) and provided pollen every two days. Bees had ad libitum access to a sucrose solution during training phases.
Individual foragers were identified during a 48-hour free-foraging period and marked with paint and numbered tags for individual-level tracking.
Experimental design
The study consisted of three experiments examining problem-solving behavior under different informational and perceptual conditions.
Pre-training
Depending on treatment, bees were given experience with:
- A visual cue (“flower”) associated with sucrose reward
- A movable object (Styrofoam ball)
- A combined interaction where bees displaced the ball to access reward
Control groups received reduced or no exposure to these elements.
Behavioral tasks
Experiment 1 (object manipulation task)
Bees were tested individually in an arena where a reward-associated “flower” was positioned on the ceiling. A ball was placed on the arena floor, and bees could only reach the flower by moving the ball underneath it and climbing onto it.
Each bee was tested once (maximum duration 15 min).
Recorded variables included:
- Task success (binary)
- Number of attempts
- Number of ball movements
- Rolling time
Experiment 2 (restricted perceptual feedback)
To assess the role of visual feedback, barriers were introduced:
- Experiment 2a: A single barrier prevented visual access to the flower from the ball’s starting position
- Experiment 2b: A multi-barrier setup required moving the ball through several compartments; control individuals were tested without a flower
Outcome variable: task success (binary).
Experiment 3 (goal-directed choice under occlusion)
Bees were tested in an arena with two visually occluded compartments. During habituation, bees observed the flower location, but during testing the flower was no longer visible from the ball’s starting position.
The task required moving the ball into the correct (previously associated) compartment.
Recorded variables included:
- Correct vs. incorrect compartment choice (binary success)
- Inspection behavior prior to ball movement
- Ball movement trajectories
- Time spent interacting with the ball
Data acquisition
Behavior was recorded using video cameras (iPhone 8) either from the side (Experiment 1) or from above (Experiments 2–3).
Movement trajectories were extracted using:
- AnimalTA (v3.2.2) for manual/assisted tracking
- DeepLabCut (v3.0.0rc9) for automated pose estimation
Tracking data were used to generate spatial trajectories and heatmaps of exploration and ball movement.
Data processing
Behavioral variables were extracted from video recordings using predefined criteria:
- Success: completion of task-specific goal (e.g., ball positioned under flower or in correct compartment)
- Attempts: directed but unsuccessful interactions toward the goal
- Moves: discrete ball displacements
- Rolling time: duration of interaction with the ball
Spatial data were processed frame-by-frame to quantify positional densities and inspection patterns. Proportional measures (e.g., inspection bias) were calculated relative to total observations per individual.
Statistical analysis
All analyses were conducted in R (RStudio).
Generalized linear mixed models (GLMMs) were used to analyze behavioral outcomes:
- Binomial models for success
- Negative binomial models for count variables
Colony identity was included as a random effect in all models to account for non-independence.
Model diagnostics (DHARMa) were used to assess dispersion, zero inflation, and residual structure. Model performance was evaluated using ROC/AUC where applicable.
Data availability
The dataset includes:
- Individual-level behavioral outcomes
- Extracted movement and inspection metrics
- Processed trajectory data
All raw and processed data, along with analysis scripts, are available in the Dryad repository.