Autonomous navigation using models from the fruit fly brain

https://flywire.ai/

This project investigates autonomous navigation in robots by leveraging computational neuroscience inspired by insect brain architectures, specifically the fruit fly (Drosophila melanogaster). Utilizing recent connectomic models, we have incorporated the optic lobes of the fruit fly for visual processing and the central complex of the sweat bee for spatial navigation (path integration). Our simulation experiments have successfully demonstrated a robot returning to a start location within a simplified square environment. However, experiments in physical environments showed limitations, particularly the inability to navigate around corners due to the lack of landmark recognition and environmental mapping.

Future directions for this research include:

1. Adding models of Mushroom Bodies to enable landmark learning and improve environmental mapping.
2. Exploring additional brain areas of the fruit fly, such as:
   • The Lateral Complex
   • The Superior Protocerebrum
   • The Inferior Protocerebrum
   • The Ventromedial area
   • The Ventrolateral area
   • The Antenna Lobes
   • The Lateral Horns
   • The Periesophageal area
3. Investigating the use of event-based (neuromorphic) cameras instead of traditional CMOS sensors to improve the efficiency and responsiveness of the visual system.
4. Extending sensory modalities beyond visual and compass inputs, for instance through olfactory sensors mimicking insect antennae, thus broadening environmental interaction capabilities.
5. Evaluating the computational efficiency of transitioning from non-spiking to spiking neural models implemented on neuromorphic hardware.

Advisor:

• Florian Röhrbein, florian.roehrbein@…

Requirements:

• Basic knowledge in Linux and Python.