Bio-inspired neural control algorithms for applications in biomimetic robotics

Abstract

Robotic systems have experienced a strong development in the last decade, due to the growing need for a robotics capability to work autonomously in non-human accessible and unstructured environments. Due to the complexity of the tasks/applications, some of the conventional robots are unable to fulfill the demands of certain tasks. Therefore, robotics designers look at how animals perform these tasks in nature to design robots with biologically inspired morphology (known as biomimetic robotics). In this approach, combining human decision making and creativity with the qualities possessed by biomimetic robots in hazardous or inaccessible environments results in an effective system capable of operating in a wide variety of environments. According to the complexity of the tasks and the redundancies of biomimetic robot designs, conventional or nonlinear control techniques have not been able to consolidate as controllers of these robots. In this last decade, the design and implementation of bio-inspired control algorithms based on the different levels of the central nervous system for biomimetic robotics have increased. Based on the above, this research work presents two motivations for the development of bio-inspired algorithms for biomimetic robotics, these motivations are 1) How the central nervous system in human/primate controls variable motion and stiffness in the fingers of a hand in an agonist-antagonist actuation system to grasp and manipulate objects in a versatile and efficient manner; and 2) How humans or animals perform autonomous navigation with obstacle avoidance efficiently in different and unstructured environments. In this direction, the main objective of this PhD Thesis is to design/model, implement and provide neurobiologically based control algorithms and a distributed neurobiologically inspired intelligent control architecture for sensorimotor control of motion, trajectory tracking and obstacle avoidance (active and reactive autonomous navigation) in unstructured environments. To achieve the main objective in this research work, two types of neurobiologically inspired algorithms have been proposed. First proposed algorithm is a corticospinal neural network that is within the field of neurophysiology and motor psychophysics for voluntary movement control and joint trajectory tracking in anthropomorphic robotic systems. This algorithm represents the levels and connectivity of the central nervous system such as cortical areas, spinal cord, and peripheral nervous system to drive agonistantagonist actuators with properties like human or primate muscles. This neurobiologically inspired algorithm was implemented to control different anthropomorphic robotic platforms (1. A two-degree-offreedom robotic finger driven by tendons and DC motors. 2. A three-degree-of-freedom robotic finger driven by tendons and shape memory alloy actuators. 3. An anthropomorphic multi-finger robotic hand, each finger having four degrees of freedom driven by tendons and DC linear motors). In this research work, we also innovated with the fabrication, design, and study of the sensing and actuating properties of a triple-layered artificial muscle based on conductive polymers, and implemented the cortico-spinal neural network for angular motion control of this biomimetic muscle. Experimental results demonstrate that the proposed corticospinal neural network implemented on different robotic platforms is efficient, reliable, stable, adaptive, and robust to different loads or perturbations, and establishing itself as a neurocontroller in anthropomorphic robotic systems for movement control and joint trajectory tracking. Second proposed algorithm is a neurobiologically inspired distributed intelligent control architecture composed of two different types of bio-inspired neural networks for autonomous navigation, obstacle avoidance and resilience capabilities in redundant biomimetic robotic systems. This architecture is based on knowledge of the active areas of human/primate motor cortex, presenting the connectivity and integration in a simplified form of premotor cortex, primary motor cortex, primary somatosensory cortex, and parietal/visual motor cortex. The bio-inspired algorithms that compose the architecture are: 1. A self-organizing direction mapping network. This bio-inspired neural network enables trajectory generation, trajectory tracking, target localization and incorporates resilience capabilities. The proposed neural network represents primary motor, premotor, primary somatosensory and visual cortex. The learning is unsupervised and uses an associative learning system to generate transformations between spatial coordinates and velocity coordinates of actuators in robotic systems. In addition, this algorithm can learn the resilience of redundant robotic systems. 2. A neural network for obstacle avoidance. This neural network is based on animal behavior and learning known as operant conditioning to control the robot's obstacle avoidance behaviors. The learning is unsupervised and the robot learns by moving in an unstructured environment full of obstacles. The distributed intelligent control architecture was implemented in terrestrial and marine mobile robotic platforms for harsh and unstructured environments. The first robotic platform is an autonomous land mobile vehicle with an electric propulsion system capable of operating for long time periods (integrating a dual system of photovoltaic panels and a methanol fuel cell for battery recharging) for observation and trajectory tracking in unstructured environments. The second robotic platform is a novel and innovative marine monitoring system called BUSCAMOS, which consists of an autonomous surface vessel combined with an unmanned autonomous underwater vehicle. The vehicles have an electric propulsion system and batteries are recharged by a dual system of solar panels and a backup diesel generator (located in the autonomous surface vehicle). Both vehicles have distributed intelligent control architecture implemented through a modular software architecture and controlled by redundant devices that provide the necessary robustness to operate in harsh and unstructured environments. Monitoring operations were performed on both land and marine robotic systems. The marine monitoring system BUSCAMOS performed long-term monitoring of oil spills (hydrocarbons), including searching for the spill and transmitting information on its location, extent, direction, and speed. Experimental results demonstrate the reliable and modular form integration, efficiency, stability, adaptive capability, resilience, and robustness of the proposed distributed intelligent control architecture for adaptive navigation and obstacle avoidance in unstructured environments for marine and terrestrial autonomous mobile robotic systems for long-term innovative applications. The neurobiologically inspired control algorithms proposed in this PhD thesis provide efficiency, reliability, stability, adaptive capability, robustness, flexibility, and resilience in redundant biomimetic robotic systems for sensorimotor control of motion, trajectory tracking, and obstacle avoidance (active and reactive autonomous navigation) in diverse and unstructured environments. This research work opens the door to know and understand in a simplified way how the levels of the central nervous system (cortical areas, spinal cord, and peripheral nervous system) interact to control in an agonistantagonist manner joint movements, as well as active and reactive autonomous spatial navigation in humans/primates. In order to design new neurobiologically inspired algorithms for redundant biomimetic robotic systems.