Predicting Drosophila Body Orientation from a Translational Trajectory using an Artificial Neural Network

Area of Science:

• Bioengineering
• Neuroscience
• Robotics

Background:

• Accurate measurement of insect body orientation is crucial for understanding flight behavior.
• Existing methods for tracking body yaw are often limited to small experimental volumes, hindering large-scale or long-duration studies.
• Translational flight trajectory data is readily available from modern tracking systems.

Purpose of the Study:

• To develop a data-driven method for predicting body yaw orientation from translational flight trajectory data.
• To enable the analysis of insect flight behavior using existing datasets that lack dedicated orientation tracking.
• To provide a practical tool for researchers studying insect navigation and behavior.

Main Methods:

• A fully connected feed-forward artificial neural network (ANN) was trained on flight trajectory and body orientation data from freely flying Drosophila.
• Input features included time-delay embeddings of ground velocity, air velocity, and inferred thrust vectors.
• Training data was augmented with random rotational transformations to improve generalization across coordinate frames.

Main Results:

• The rotation-augmented ANN model accurately predicted body yaw orientation from translational flight data.
• The model achieved a median mean absolute angular error of 10.51° on a test set of 3,313 trajectories.
• Accurate heading recovery was demonstrated across the full [-π, π) range.

Conclusions:

• The developed estimator offers a practical solution for recovering body orientation from existing trajectory datasets.
• This approach significantly expands the scope of behavioral and computational analyses of insect navigation.
• It overcomes limitations of specialized hardware, making large-scale insect flight studies more accessible.