Roboticists reverse engineer zebrafish navigation

(Press-News.org) A paradox of neuroscience is that while brains evolve within specific sensory and physical environments, neural circuits are usually studied in isolation under controlled laboratory conditions. But we can’t fully understand how environmental factors shape brain function without considering the body in which that brain evolved.

The BioRobotics Lab in EPFL’s School of Engineering specializes in developing bioinspired robots to tease apart the brain-body interactions involved in sensorimotor coordination. Now, they have published a study in Science Robotics that provides detailed insight into embodiment, or how the body affects perception, in larval zebrafish.

“Our simulated larval zebrafish provided virtual embodiment, which allowed us to observe its reaction to simulated fluid dynamics and visual scenes. Then, we used a physical robot to observe these interactions in the real world. These connections to the environment can’t be studied with an isolated brain in a lab,” summarizes BioRobotics Lab head Auke Ijspeert.

A fish-eye view

With their translucent bodies, tiny larval zebrafish offer optical access to all their neurons, making them well-studied animal models in biomedical research. For the study, neurobiologist Eva Naumann and her team at Duke University provided a neural network architecture derived from real-time imaging data from the brains of live zebrafish. They also tracked visually driven behavior of the tiny fish, and recorded their reactions when presented with varied visual stimuli that mimicked what they might encounter in flowing water.

The BioRobotics Lab then worked with Naumann and her team to develop a simulation that faithfully reproduced zebrafish visual processing, body mechanics, and neural circuits, from retina to spinal cord. In experiments targeting the optomotor response – the reflexive swimming that helps fish compensate for water currents – the virtual animal closely replicated the behavior of real larval zebrafish.

“It was exciting that we replicated all the different behaviors that Eva and her team observed in the live fish – it suggests we succeeded in reverse engineering the circuitry,” Ijspeert says.

In the process, the team discovered that most neural signals driving zebrafish behavior come from a relatively small part of the retina. Their simulation even predicted two previously unidentified neuron types that explained the live fish’s response to unusual stimuli.

To further validate their work, EPFL postdoctoral researcher Xiangxiao Liu built an 80-cm robotic zebrafish larva equipped with two cameras for eyes, motors to move its tail segments, and the same neural circuits as the simulated zebrafish. In experiments in Lausanne’s Chamberonne River, the robot was able to keep from being swept downstream, even in the disorder of a natural environment.

“The emergence of the optomotor response from our neural circuitry is significant, as some of an animal’s response to any stimulus is random. Despite this randomness, the neural circuitry still converged to reorient the robot and maintain its position,” Liu says.

An open-source platform for studying animal behavior

The BioRobotics Lab is already extending this research to study zebrafish swimming patterns. Their simulation and robot design are also available as open-source tools for researchers to study visuomotor coordination in zebrafish and in other animals. Indeed, Ijspeert emphasizes that the work demonstrates how crucial models and simulations are for understanding which sensorimotor mechanisms are sufficient for certain biological functions.

“In animal experiments, you can only show which sensorimotor mechanisms are necessary to function, but not which are sufficient, but because you can’t remove all mechanisms except one in animals. Here, we have shown that vision alone is in principle sufficient for zebrafish to maintain their position, which is a challenging and non-trivial result.”

END