Whole-body physics simulation of fruit fly locomotion

Area of Science:

• Computational Neuroscience
• Biophysics
• Robotics

Background:

• The nervous system's control over sensorimotor behavior is intrinsically linked to an animal's physical body.
• Accurate modeling of neural control necessitates detailed biomechanical representations.
• Simulating complex behaviors requires integrating body dynamics with neural control mechanisms.

Purpose of the Study:

• To introduce a versatile, whole-body physics-based model of the fruit fly Drosophila melanogaster.
• To enable the simulation and study of diverse fly behaviors, including terrestrial and aerial locomotion.
• To develop a platform for investigating the neural control of sensorimotor behavior in an embodied context.

Main Methods:

• Development of an anatomically detailed, whole-body biomechanical model of Drosophila melanogaster within a physics simulator.
• Implementation of phenomenological models for fluid and adhesion forces to support locomotion.
• Application of data-driven, end-to-end reinforcement learning to train neural network controllers for naturalistic locomotion and visually guided flight.

Main Results:

• Successful replication of realistic walking and flight behaviors, validating the model's versatility.
• Training of neural network controllers capable of generating naturalistic locomotion along complex trajectories.
• Demonstration of visually guided flight using a hierarchical motor control system with visual sensors.

Conclusions:

• The developed whole-body fruit fly model provides a powerful, open-source platform for studying embodied sensorimotor control.
• The framework supports diverse behaviors and facilitates research into neural control mechanisms.
• This approach advances our understanding of how physical bodies shape neural control of behavior.