Skip to main content
Dryad logo

Impacts of leg, wing, and body collisions on flight performance in carpenter bees


Burnett, Nicholas; Combes, Stacey (2022), Impacts of leg, wing, and body collisions on flight performance in carpenter bees, Dryad, Dataset,


Flying insects often forage among cluttered vegetation that forms a series of obstacles in their flight path. Recent studies have focused on behaviors needed to navigate clutter while avoiding all physical contact, and as a result, we know little about flight behaviors that do involve encounters with obstacles. Here, we challenged carpenter bees (Xylocopa varipuncta) to fly through narrow gaps in an obstacle course to determine the kinds of obstacle encounters they experience, as well as the consequences for flight performance. We observed three kinds of encounters: leg, body, and wing collisions. Wing collisions occurred most frequently (in about 40% of flights, up to 25 times per flight) but these had little effect on flight speed or body orientation. In contrast, body and leg collisions, which each occurred in about 20% of flights (1-2 times per flight), resulted in decreased flight speeds and increased rates of body rotation (yaw). Wing and body collisions, but not leg collisions, were more likely to occur in wind versus still air. Thus, physical encounters with obstacles may be a frequent occurrence for insects flying in some environments, and the immediate effects of these encounters on flight performance depend on the body part involved.


Freely flying female carpenter bees (Xylocopa varipuncta, n = 15) were collected from the University of California, Davis campus and subjected to flight challenges in a laboratory flight tunnel. Individual bees were placed in the flight tunnel (20 x 19 x 115 cm; width x height x length), which included a series of vertical columns (hereafter ‘obstacles’) that spanned the middle of the tunnel (obstacle diameter = 7 mm, space between obstacles = 34.44 ± 2.80 mm; mean ± SD). The bees’ wing spans (45.19 ± 2.11 mm, from tip to tip) were larger than the size of the gaps between obstacles; thus, bees could not fly straight through gaps, but instead needed to rotate their body (e.g., yaw) in order to pass through. Obstacles were attached to a mechanical arm that oscillated laterally (amplitude = 21 mm, frequency = 2 Hz) or remained stationary. Fans at each end of the tunnel could be turned on to produce a gentle breeze (mean velocity = 0.54 m/s) or off for still air. Wind direction was constant: bees flying in one direction experienced headwinds and in the other direction tailwinds. Up to 12 flights through the obstacles were elicited from each bee, using full spectrum lights that were alternately turned on and off at each end of the tunnel. Obstacle motion (stationary versus oscillating) was fixed for a given bee, but all bees experienced both wind and still air, with the order of the wind condition switched after approximately six flights. Thus, four different flight conditions were tested on the group of bees: still air with stationary obstacles (n = 40 flights), still air with oscillating obstacles (n = 34), wind with stationary obstacles (n = 42), and wind with oscillating obstacles (n = 29).

Flights were filmed with two synchronized Phantom v611 cameras (Vision Research, Inc., Wayne, NJ, USA) sampling at 1500 frames/s and positioned 30º from the vertical on opposite sides of the obstacles. Cameras were calibrated using a standard checkerboard calibration method and built-in MATLAB functions.

In each video, the positions of the bee’s head (midpoint between antennae), thorax (approximating the body centroid), and wing tips were tracked with the machine-learning software DeepLabCut. Tracked points were checked and manually corrected, and obstacle positions were labeled using DLTdv6 in MATLAB. Labeled positions were converted from two-dimensional coordinates in each camera view into three-dimensional space using built-in MATLAB functions.

We classified and counted each physical encounter that occurred between the bees and obstacles. The most common encounters were body collisions (the head, thorax, or abdomen contacted the obstacles), leg collisions (one or more forelegs contacted the obstacles), and wing collisions (the distal half of one or more forewings contacted the obstacles).


We next assessed how encounters with obstacles affected flight performance. In each video, we identified the first occurrence of each body, leg, and wing collision, and defined a 20-ms period before and after each encounter. This temporal window allows us to quantify performance immediately before and after obstacle encounters, as in, to determine how flight performance changes during encounters. Some of these pre- and post-encounter periods contained additional collisions with obstacles, but they were included in the analysis to reflect the true nature of flight in clutter (i.e., flights with just one obstacle encounter were rare). From this analysis, videos yielded either a single encounter type (n = 42 flights), two encounter types (n = 26), or all three encounter types (n = 9).

For each encounter, we measured the change in horizontal ground speed and yaw between the pre- and post-collision periods, as well as the post-collision yaw rate, where yaw was the body angle about the vertical axis. To calculate kinematics, we smoothed the three-dimensional position data with cubic smoothing spline curves via the ‘smooth.spline’ function in the R package stats. Horizontal ground speed was calculated as the change in x-y position (lateral and longitudinal motions, omitting vertical motion) per time. Yaw was calculated by converting the Cartesian coordinates of the head and thorax to spherical coordinates via the ‘cart2sph’ function in the R package pracma and finding the horizontal angle between the two body points and the long axis of the tunnel. The yaw rate was calculated over the 20 ms post-collision by taking the derivative of yaw with respect to time.


National Science Foundation, Award: 1711980