General purpose robot end-effectors are fast, accurate and on occasions flexible, but inherently narrow in the tasks they perform. In endeavouring to mimic the human hand (the most dexterous end-effector), a tele-operated 5 fingered, 23 degree of freedom (DOF) dexterous anthropomorphic robotic hand has been designed, constructed and tested. This hand is biomimetically actuated by 20 braided pneumatic Muscle Actuators (pMAs). The thesis shows the structured development of this complex anthropomorphic robotics structure, a method of taking advantage of the inherent nature of the pMAs which weakens linearly as it contracts, a system of transmitting power from pMAs to the joints in the robotic hand, a novel method of demonstrating the dexterity and speed of the robotic hand and development of a micro-controller (μC) based pMA controller.

Object recognition is a very challenging problem particularly for mobile robot applications such as service robots. The challenges are due to the variation of the point of view and imaging conditions which may involve occlusion of the object. The suitable approach in our case is the local features-based approach. The methods used in this approach include two stages; the first is the extraction of local features (also called interest points), and the second is the matching operation stage. Scale Invariant Feature Transform (SIFT) is recognised as one of the most robust and efficient algorithms for the extraction of local features. We adopt the SIFT algorithm for the determination of the keypoints used in the matching stage for object detection and pose estimation. The focus of our work is mainly on the matching stage, analysing the different techniques used, and proposing some improvements, taking into consideration the balance between reliability and efficiency required in mobile robot object recognition system. Both sources of information on the disposition of the keypoints are used in the matching operation; the relative positions of the keypoints in the descriptors space and the geometric information about their positions in the image. We use a simplified method of object recognition and pose estimation that is based on robust selection of matching keypoints exploiting in addition to the descriptors matching the geometric information on the keypoints. Clusters of reliably selected matched keypoints are used in object pose estimation using the Least Squares Method. Analysis and experimental results on different points of the matching operation are presented.

In this research, commercially available Nickel-Titanium based shape memory alloy actuators are used to design and develop a novel bundled actuator. The bundled actuators are then used in a special configuration to actuate and control a low cost one degree of freedom (dof) anthropomorphic finger. SMAs possess hysteretic and nonlinear behaviour when heated (actuated) and cooled (relaxed). Hence, a full understanding of the behaviour of these actuators is essential to control them. In order to characterise the behaviour of the actuators, an instrumented test rig was designed and developed to characterise the commercially available Nickel-Titanium SMA actuators obtained from the two manufacturers. The test rig was equipped with the laser displacement sensor, the load cell and the thermocouple. All the measurements were taken by a PC based data acquisition system. Nickel-Titanium based SMA actuators (Flexinol® and Biometal'") obtained from the two manufacturers were characterised using the test rig. The Flexinol'" actuator was selected for the use in the bundled actuator on the basis of better drift property. The existing bundle designs are merely a parallel arrangement of individual actuators crimped or screwed to the end plates on both sides. However, this may cause uneven strains (and forces) in individual actuators of the bundle. The novel bundle proposed in this research used a single 960 mm long actuator looped around frictionless supports to form a 12-string bundled actuator. Design and simulation of a one dof finger was carried out by using a set of three bundled actuators arranged in a special configuration. This made it possible to use a simple control to achieve seven possible for the proposed finger.

A fundamental question in biology is how cooperation and especially altruism evolves in nature. There is debate as to whether altruism evolves by individual or group selection and if kin selection is the key explanation. Recent research in evolutionary robotics has provided some evidence that altruism emerges in populations of evolving robot controllers. A set of experiments is presented here that evolve heterogeneous, but related, populations of simulated abstract robot controllers with a biologically inspired implicit fitness function based simply on 'eating' enough to survive. In predator-prey simulations, an internal fitness function that is implicit in the selection of individuals is implemented. Predators are paired in a series of trials where they interact to 'consume' the prey and those that survive are selected to reproduce and proceed to the next generation. When individual level selection is implemented, true artificial altruism emerges naturally, where an individual sacrifices its own 'life' so as to allow a related robot to survive. The altruistic trait evolves in situations where there is a high probability that neither of a pair would be successful if both attempt to survive. When situated communication is available, it is used by a robot to claim the prey once located, signalling to the other robot to behave altruistically. Under 'breeding pair' level selection where a pair of robots are required to survive together, they cooperate to ensure that both are successful and when situated communication is available, it is used by the robots to signal that the prey has been located, increasing the amount of collaboration in the population. In selecting for competition, where the robots are not allowed to survive as a pair, in addition to competing, the robots also act altruistically to achieve the goal of only one survivor.

In mobile robotics, visual perception in unknown environments consists mainly of two tasks: generation of a map and localization of the agent within it, using images as input. This problem is commonly called Visual Simultaneous Localization and Mapping ( Visual SLA M). This project aims of increasing the capacities of spatial reasoning in robotics systems based on cognitive vision. Baycsian formulation of visual mapping is extended to consider geometric properties for increasing of scene understand- ing. In this search, this work presents a framework that covers the computa- tional and algorithmic perspectives in Cognitive Vision. proposed by Marr. A part of the work is devoted to develop the representational framework for describing the visual phenomenon in monocular SLAM, unifying concepts about the existent parametrization and extending the analysis to geometric properties and relations in the scene. This yields a novel strategy for aug- mented mapping with high lcvellandmarks based on planar surfaces. After, we explore the computational level of visual SLAM through a sym- bolic framework for describing the elements required for spatial reasoning, involved in visual SLAM. An ontology for spatial reasoning is proposed upon visual SLAM context. An axiomatic set in visual scenario is related to prop- erties in projective geometry, assuring a scheme for semantic attachment. Finally, a novel approach to solve visual SLAM based on spatial reasoning is presented, focusing on the algorithmic level. The model of latent geomet- ric constraints is presented as a non-parametric Baycsian extension of visual SLAM. The generative model produces multiples scenarios with different visual conditions. Although this thesis is focused on solving visual SLAM, the proposed ap- proach can be conceived as a methodology for the design of cognitive dynamic systems.

No description available.

No description available.

No description available.

No description available.

No description available.