This dataset contains data analyzed from videos of moths feeding from artificial flowers and has analysis files for both moth learning (Fig 2) and proboscis sweeping (Fig 3 and 4). Organization of files and folders DLCdataSet dlcTrainingData Each folder contains 1) images (*.png) from each video and 2) the corresponding annotation files (*.csv) for 7 different bodyparts (Butt, flowerCenter, head, LeftAntennae, RightAntennae, proboscisJoint and ProboscisTip). DLCdataSet Output contains a folder titled "EntireDataSet" that contains *.h5 and *.pickle files exported by DeepLabCut tracking for all the videos in the dataset. For each bodypart there is a x,y position and a likelihood parameter. MothLearning_dataFolders DLCtrackingResults Files include *.csv that were used to compare the learning data from a subset of videos that were manually annoted and dlc-tracked. These also contain manual annotations that were used to build the SVM classifier for classifying true/false visits. MothLearning_dataFolders Output contains folders titled Step*_* that follow directly from the name of codes in github respository: https://github.com/TanviDeora/MothLearning. These files contain 1) *.csv files that specify x,y position of moth centroid computed from videos and 2) *.mat files from synchronized motion sensor in the flower nectary. These folders contain all the data needed to create Fig. 2 and the associated statistics. ProboscisTracking_dataFolders CircleParameters contains 1) a single video frame (*.png) with the overhead view of the flower for all videos included in the dataset and 2) a file titled "circle_parameters.csv" that has x,y location and the flower radius for each video. These files are used to transform the probsocis tracks from global to flower centric coordinates as well as convert x,y pixel position to cm. ProboscisTracking_dataFolders v3 This folders contains all the analysis output using the codes in the repository - https://github.com/TanviDeora/ProboscisTracking. The folders are: 1. Raw tracks - *.csv files for each moth containing raw tracks (x,y position) of proboscis tip as exported by the DeepLabCut for all relevant visits. 2. Fitered tracks - *.csv files for each moth containing filtered tracks of proboscis tip for all relevant visits, incuding the ones that were manually tracked. The data in the folder titled "FinalCompiledTracks" contains the x,y positions for all visits that were used for all downstream analysis. 3. OtherBodyparts - *.csv files containing x,y position of other body parts. The first visit for only a subset of moth videos were analyzed. Files includes 1.*mothName*_*bodyPart*.csv for each bodyPart that was analyzed and 2.*mothName*_probHeadAngle.csv and 3.*mothName*_ProbHeadAntennae.csv. 4. RadiusAndAngle - *.csv files for each moth containing x, y position for the proboscis tip in the flower centric coordinates as well as the relative radial angle and relative radial position. These are used to generate Fig.3 and Fig. 4 for the associated paper. 5. SweepDynamics*.csv files contains different parameters of proboscis sweeping. 

The interaction between insects and the flowers they pollinate has driven the evolutionary diversity of both insects and flowering plants, two groups with the most numerous species on Earth. Insects use vision and olfaction to localize host plants, but we know relatively little about how they find the tiny nectary opening in the flower, which can be well beyond their visual resolution. Especially when vision is limited, touch becomes crucial in successful insect–plant pollination interactions. Here, we studied the remarkable feeding behavior of crepuscular hawkmoths Manduca sexta, which use their long, actively controlled, proboscis to expertly explore flower-like surfaces. Using machine vision and 3D-printed artificial flower-like feeders, we revealed a novel behavior that shows moths actively probe surfaces, sweeping their proboscis from the feeder edge to its center repeatedly until they locate the nectary opening. Moreover, naive moths rapidly learn to exploit these flowers, and they adopt a tactile search strategy to more directly locate the nectary opening in as few as three to five consecutive visits. Our results highlight the proboscis as a unique active sensory structure and emphasize the central role of touch in nectar foraging insect–plant pollinator interactions.