Introduction - NeLy-EPFL/cobar-miniproject-2023 GitHub Wiki

Scientific Background

Animals are capable of highly dexterous and adaptive behaviors unmatched by even the most advanced robots. However, mechanisms underlying such agile body coordination remain largely obscure. Understanding these mechanisms can both improve our fundamental understanding of neurobiology and facilitate advances in robotics and artificial intelligence.

We hope to probe into these principles by reverse-engineering sensorimotor control in Drosophila melanogaster, the fruit fly. Drosophila is a widely used animal model in biological research that benefits from being highly genetically tractable, a feature that allows researchers to target and study specific neurons repeatedly. With a smaller nervous system (~100k neurons), the fly nervous system may also be easier to model in its entirety.

In this course and mini-project, you will develop an artificial controller to control the behavior of a virtual fly in a physics simulator (MuJoCo). The simulated fly has an accurate body morphology that was reconstructed from a micro CT scan of a real animal. For more details about this fly model, you can read Lobato-Rios et al, 2022. Controlling the behavior of a simulated fly serves as an intermediate step toward the ultimate goal of building a fly-inspired robot.

Recall that this course is titled “Controlling Behavior in Animals and Robots.” We hope that this miniproject will provide you with a hands-on experience in applying biologically-inspired control principles to artificial systems.

Miniproject Logistics

Task: Through discussions with your partner and the course instructors you will design a neural controller that transforms sensory signals (e.g., vision, olfaction, touch...) into appropriate actions (jump, walk, turn, grooming...).

Groups: You will work in groups of 2-3. You can suggest teammates that you want to work with, but we will ultimately assign groups to ensure a balanced background within each group.

Class organization: The second half of the course (after the midterm exam) is entirely dedicated to this miniproject. You will work in groups to carry out the project.

Evaluation: Each group will hand in the code, a written report, and give a presentation on their project. The code, the report, and the presentation will be evaluated holistically and together will account for 40% of your final grade. We are excited to see what you can accomplish in building a first whole organism neuromechanical simulation!