Simulation of Salamander Locomotion

Abstract

As part of the research on locomotion controller that aims to produce robots whose design is inspired by Nature, this thesis intends to develop a simulator of the salamander locomotion. It investigates, in particular, what type of circuitry can produce and modulate the neural activity for swimming and trotting. In water, the animal moves by propagating a traveling wave of muscular contractions along the body while holding the limbs against it in an undulatory gait. On the ground, instead, the salamander switches into a stepping gait with the body making a S-shaped standing wave coordinated with the movements of the limbs. While the connectivity and properties of the neural circuit that regulate the movement of the salamander have not been decoded so far, the locomotor controller that is used here corresponds to the general organization hypothesis of neurobiologists who believe the controller is made of a lamprey-like Central Pattern Generator body (CPG) and two limbs CPGs. The neural circuit is implemented as a Dynamic Artificial Neural Network whose units, modeled as leaky integrators, try to simulate accurately the properties of similar biological neurons. A three-dimensional mechanical model is developed that also takes into account the realistic forces acting upon it in order to verify whether the locomotor controller is able to reproduce the desired pattern for both swimming and trotting. The neural model controls the mechanical one through the contractions of the muscles. In an incremental approach, a Genetic Algorithm is used to develop the neural circuit capable of reproducing the salamander locomotion in the mechanical simulation of the animal. The connections of the circuit and the properties of the neurons are evolved in three stages. First, the singular segmental oscillator is evolved. Second, the evolution of the intersegmental connections in the body of the salamander, composed of forty identical segment oscillators, follows. Third, the limb CPG and its coupling with the body CPG is finally developed. The neural controller resulting from these three steps is able to switch from one gait to the other, to modulate the velocity, and to change the direction based on the tonic input received from the brainstem. The locomotor circuit is also capable of replicating in the mechanical body the gaits observed in the real salamander, which, considering the intention of the thesis, is proving the successful simulation of the animal locomotion.