Neuromechanical Simulations of Animats with Decentralised Control
The transition from aquatic to terrestrial environments represents a significant event in the history of evolution. For this transition to occur, animals had to adapt their morphology, physiology and locomotory skills to handle the challenges and interactions they would face in terrestrial environments. At the level of the neural circuits, the intrinsic control mechanisms which permitted this transition remain a mystery, but could have been hypothetically achieved using a combination of simple control rules related to Central Pattern Generators (CPGs) and sensory feedback. In this thesis, we investigate decentralised control and how sensory feedback could modulate CPGs to provide gait coordination as well as gait transitions when the body is exposed to different environmental media. In particular, we make extensive use of neuromechanical simulations using networks of simplified oscillators to model the dynamical properties of the spinal cord in a set of morphologies when tested over a range of locomotory tasks. To achieve this, we developed and present a novel Framework for Animals and Robots Modelling and Simulation (FARMS) for running neuromechanical simulations. We explore how simple circuits can achieve locomotion for a range of animals and robots in various environments. We also propose a method for determining sensor-motor maps by morphological probing using Hebbian learning, which resulted in emergent adaptive gaits when combined with an oscillator-based controller. We present a neuromechanical model of drosophila melanogaster, NeuroMechFly, to replicate and assist the animal experiments in future studies and used decentralised control to reproduce walking behaviour. Finally, we propose and characterise reduced decentralised and decoupled models of neural circuits, capable of reproducing amphibious locomotion for polypterus, salamander and centipede in a series of locomotory tasks.
EPFL_TH9245.pdf
n/a
openaccess
copyright
52.36 MB
Adobe PDF
d484517378fc004e7528ee6a89569e23