Fok Lo Ming. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2002. / Includes bibliographical references (leaves 133-135). / Abstracts in English and Chinese. / Chapter 1. --- Introduction --- p.1 / Chapter 1.1 --- Multi-fingered Robotic Hands --- p.1 / Chapter 1.2 --- Grasping Force Optimization --- p.2 / Chapter 1.3 --- Neural Networks --- p.6 / Chapter 1.4 --- Previous Work for Grasping Force Optimization --- p.9 / Chapter 1.5 --- Contributions of this work --- p.10 / Chapter 1.6 --- Organization of this thesis --- p.12 / Chapter 2. --- Problem Formulations --- p.13 / Chapter 2.1 --- Grasping Force Optimization without Joint Torque Limits --- p.14 / Chapter 2.1.1 --- Linearized Friction Cone Approach --- p.15 / Chapter i. --- Linear Formulation --- p.17 / Chapter ii. --- Quadratic Formulation --- p.18 / Chapter 2.1.2 --- Nonlinear Friction Cone as Positive Semidefinite Matrix --- p.19 / Chapter 2.1.3 --- Constrained Optimization with Nonlinear Inequality Constraint --- p.20 / Chapter 2.2 --- Grasping Force Optimization with Joint Torque Limits --- p.21 / Chapter 2.2.1 --- Linearized Friction Cone Approach --- p.23 / Chapter 2.2.2 --- Constrained Optimization with Nonlinear Inequality Constraint --- p.23 / Chapter 2.3 --- Grasping Force Optimization with Time-varying External Wrench --- p.24 / Chapter 2.3.1 --- Linearized Friction Cone Approach --- p.25 / Chapter 2.3.2 --- Nonlinear Friction Cone as Positive Semidefinite Matrix --- p.25 / Chapter 2.3.3 --- Constrained Optimization with Nonlinear Inequality Constraint --- p.26 / Chapter 3. --- Recurrent Neural Network Models --- p.27 / Chapter 3.1 --- Networks for Grasping Force Optimization without Joint Torque Limits / Chapter 3.1.1 --- The Primal-dual Network for Linear Programming --- p.29 / Chapter 3.1.2 --- The Deterministic Annealing Network for Linear Programming --- p.32 / Chapter 3.1.3 --- The Primal-dual Network for Quadratic Programming --- p.34 / Chapter 3.1.4 --- The Dual Network --- p.35 / Chapter 3.1.5 --- The Deterministic Annealing Network --- p.39 / Chapter 3.1.6 --- The Novel Network --- p.41 / Chapter 3.2 --- Networks for Grasping Force Optimization with Joint Torque Limits / Chapter 3.2.1 --- The Dual Network --- p.43 / Chapter 3.2.2 --- The Novel Network --- p.45 / Chapter 3.3 --- Networks for Grasping Force Optimization with Time-varying External Wrench / Chapter 3.3.1 --- The Primal-dual Network for Quadratic Programming --- p.48 / Chapter 3.3.2 --- The Deterministic Annealing Network --- p.50 / Chapter 3.3.3 --- The Novel Network --- p.52 / Chapter 4. --- Simulation Results --- p.54 / Chapter 4.1 --- Three-finger Grasping Example of Grasping Force Optimization without Joint Torque Limits --- p.54 / Chapter 4.1.1 --- The Primal-dual Network for Linear Programming --- p.57 / Chapter 4.1.2 --- The Deterministic Annealing Network for Linear Programming --- p.59 / Chapter 4.1.3 --- The Primal-dual Network for Quadratic Programming --- p.61 / Chapter 4.1.4 --- The Dual Network --- p.63 / Chapter 4.1.5 --- The Deterministic Annealing Network --- p.65 / Chapter 4.1.6 --- The Novel Network --- p.57 / Chapter 4.1.7 --- Network Complexity Analysis --- p.59 / Chapter 4.2 --- Four-finger Grasping Example of Grasping Force Optimization without Joint Torque Limits --- p.73 / Chapter 4.2.1 --- The Primal-dual Network for Linear Programming --- p.75 / Chapter 4.2.2 --- The Deterministic Annealing Network for Linear Programming --- p.77 / Chapter 4.2.3 --- The Primal-dual Network for Quadratic Programming --- p.79 / Chapter 4.2.4 --- The Dual Network --- p.81 / Chapter 4.2.5 --- The Deterministic Annealing Network --- p.83 / Chapter 4.2.6 --- The Novel Network --- p.85 / Chapter 4.2.7 --- Network Complexity Analysis --- p.87 / Chapter 4.3 --- Three-finger Grasping Example of Grasping Force Optimization with Joint Torque Limits --- p.90 / Chapter 4.3.1 --- The Dual Network --- p.93 / Chapter 4.3.2 --- The Novel Network --- p.95 / Chapter 4.3.3 --- Network Complexity Analysis --- p.97 / Chapter 4.4 --- Three-finger Grasping Example of Grasping Force Optimization with Time-varying External Wrench --- p.99 / Chapter 4.4.1 --- The Primal-dual Network for Quadratic Programming --- p.101 / Chapter 4.4.2 --- The Deterministic Annealing Network --- p.103 / Chapter 4.4.3 --- The Novel Network --- p.105 / Chapter 4.4.4 --- Network Complexity Analysis --- p.107 / Chapter 4.5 --- Four-finger Grasping Example of Grasping Force Optimization with Time-varying External Wrench --- p.109 / Chapter 4.5.1 --- The Primal-dual Network for Quadratic Programming --- p.111 / Chapter 4.5.2 --- The Deterministic Annealing Network --- p.113 / Chapter 4.5.3 --- The Novel Network --- p.115 / Chapter 5.5.4 --- Network Complexity Analysis --- p.117 / Chapter 4.6 --- Four-finger Grasping Example of Grasping Force Optimization with Nonlinear Velocity Variation --- p.119 / Chapter 4.5.1 --- The Primal-dual Network for Quadratic Programming --- p.121 / Chapter 4.5.2 --- The Deterministic Annealing Network --- p.123 / Chapter 4.5.3 --- The Novel Network --- p.125 / Chapter 5.5.4 --- Network Complexity Analysis --- p.127 / Chapter 5. --- Conclusions and Future Work --- p.129 / Publications --- p.132 / Bibliography --- p.133 / Appendix --- p.136

Les robots sont de plus en plus utilisés dans un cadre social. Il ne suffit plusde partager l’espace avec des humains, mais aussi d’interagir avec eux. Dansce cadre, il est attendu du robot qu’il comprenne un certain nombre de signauxambiguës, verbaux et visuels, nécessaires à une interaction humaine. En particulier, on peut extraire beaucoup d’information, à la fois sur l’état d’esprit despersonnes et sur la dynamique de groupe à l’œuvre, en connaissant qui ou quoichaque personne regarde. On parle de la Cible d’attention visuelle, désignéepar l’acronyme anglais VFOA. Dans cette thèse, nous nous intéressons auxdonnées perçues par un robot humanoı̈de qui participe activement à une in-teraction sociale, et à leur utilisation pour deviner ce que chaque personneregarde.D’une part, le robot doit “regarder les gens”, à savoir orienter sa tête(et donc la caméra) pour obtenir des images des personnes présentes. Nousprésentons une méthode originale d’apprentissage par renforcement pourcontrôler la direction du regard d’un robot. Cette méthode utilise des réseauxde neurones récurrents. Le robot s’entraı̂ne en autonomie à déplacer sa tête enfonction des données visuelles et auditives. Il atteint une stratégie efficace, quilui permet de cibler des groupes de personnes dans un environnement évolutif.D’autre part, les images du robot peuvent être utilisée pour estimer lesVFOAs au cours du temps. Pour chaque visage visible, nous calculons laposture 3D de la tête (position et orientation dans l’espace) car très fortementcorrélée avec la direction du regard. Nous l’utilisons dans deux applications.Premièrement, nous remarquons que les gens peuvent regarder des objets quine sont pas visible depuis le point de vue du robot. Sous l’hypothèse quelesdits objets soient regardés au moins une partie du temps, nous souhaitonsestimer leurs positions exclusivement à partir de la direction du regard despersonnes visibles. Nous utilisons une représentation sous forme de carte dechaleur. Nous avons élaboré et entraı̂né plusieurs réseaux de convolutions afinde d’estimer la régression entre une séquence de postures des têtes, et les posi-tions des objets. Dans un second temps, les positions des objets d’intérêt, pou-vant être ciblés, sont supposées connues. Nous présentons alors un modèleprobabiliste, suggéré par des résultats en psychophysique, afin de modéliserla relation entre les postures des têtes, les positions des objets, la directiondu regard et les VFOAs. La formulation utilise un modèle markovien à dy-namiques multiples. En appliquant une approches bayésienne, nous obtenonsun algorithme pour calculer les VFOAs au fur et à mesure, et une méthodepour estimer les paramètres du modèle.Nos contributions reposent sur la possibilité d’utiliser des données, afind’exploiter des approches d’apprentissage automatique. Toutes nos méthodessont validées sur des jeu de données disponibles publiquement. De plus, lagénération de scénarios synthétiques permet d’agrandir à volonté la quantitéde données disponibles; les méthodes pour simuler ces données sont explicite-ment détaillée. / Robots are more and more used in a social context. They are required notonly to share physical space with humans but also to interact with them. Inthis context, the robot is expected to understand some verbal and non-verbalambiguous cues, constantly used in a natural human interaction. In particular,knowing who or what people are looking at is a very valuable information tounderstand each individual mental state as well as the interaction dynamics. Itis called Visual Focus of Attention or VFOA. In this thesis, we are interestedin using the inputs from an active humanoid robot – participating in a socialinteraction – to estimate who is looking at whom or what.On the one hand, we want the robot to look at people, so it can extractmeaningful visual information from its video camera. We propose a novelreinforcement learning method for robotic gaze control. The model is basedon a recurrent neural network architecture. The robot autonomously learns astrategy for moving its head (and camera) using audio-visual inputs. It is ableto focus on groups of people in a changing environment.On the other hand, information from the video camera images are used toinfer the VFOAs of people along time. We estimate the 3D head poses (lo-cation and orientation) for each face, as it is highly correlated with the gazedirection. We use it in two tasks. First, we note that objects may be lookedat while not being visible from the robot point of view. Under the assump-tion that objects of interest are being looked at, we propose to estimate theirlocations relying solely on the gaze direction of visible people. We formulatean ad hoc spatial representation based on probability heat-maps. We designseveral convolutional neural network models and train them to perform a re-gression from the space of head poses to the space of object locations. Thisprovide a set of object locations from a sequence of head poses. Second, wesuppose that the location of objects of interest are known. In this context, weintroduce a Bayesian probabilistic model, inspired from psychophysics, thatdescribes the dependency between head poses, object locations, eye-gaze di-rections, and VFOAs, along time. The formulation is based on a switchingstate-space Markov model. A specific filtering procedure is detailed to inferthe VFOAs, as well as an adapted training algorithm.The proposed contributions use data-driven approaches, and are addressedwithin the context of machine learning. All methods have been tested on pub-licly available datasets. Some training procedures additionally require to sim-ulate synthetic scenarios; the generation process is then explicitly detailed.

Mobile robots for rough terrain are of interest to researchers as their range of possible uses is large, including exploration activities for inhospitable areas on Earth and on other planets and bodies in the solar system, searching in disaster sites for survivors, and performing surveillance for military applications. Nature generally achieves land movement by walking using legs, but additional modes such as climbing, jumping and rolling are all produced from legs as well. Robotics tends not to use this integrated approach and adds additional mechanisms to achieve additional movements. The spherical device described within this thesis, called Jollbot, integrated a rolling motion for faster movement over smoother terrain, with a jumping movement for rougher environments. Jollbot was developed over three prototypes. The first achieved pause-and-leap style jumps by slowly storing strain energy within the metal elements of a spherical structure using an internal mechanism to deform the sphere. A jump was produced when this stored energy was rapidly released. The second prototype achieved greater jump heights using a similar structure, and added direction control to each jump by moving its centre of gravity around the polar axis of the sphere. The final prototype successfully combined rolling (at a speed of 0.7 m/s, up 4° slopes, and over 44 mm obstacles) and jumping (0.5 m cleared height), both with direction control, using a 0.6 m spherical spring steel structure. Rolling was achieved by moving the centre of gravity outside of the sphere’s contact area with the ground. Jumping was achieved by deflecting the sphere in a similar method to the first and second prototypes, but through a larger percentage deflection. An evaluation of existing rough terrain robots is made possible through the development of a five-step scoring system that produces a single numerical performance score. The system is used to evaluate the performance of Jollbot.

Many environments are inaccessible or hazardous for humans. Remaining debris after earthquake and fire, ship hulls, bridge installations, and oil rigs are some examples. For these environments, major effort is being placed into replacing humans with robots for manipulation purposes such as search and rescue, inspection, repair, and maintenance. Mobility, manipulability, and stability are the basic needs for a robot to traverse, maneuver, and manipulate in such irregular and highly obstructed terrain. Hexapod walking robots are as a salient solution because of their extra degrees of mobility, compared to mobile wheeled robots. However, it is essential for any multi-legged walking robot to maintain its stability over the terrain or under external stimuli. For manipulation purposes, the robot must also have a sufficient workspace to satisfy the required manipulability. Therefore, analysis of both workspace and stability becomes very important. An accurate and concise inverse kinematic solution for multi-legged robots is developed and validated. The closed-form solution of lateral and spatial reachable workspace of axially symmetric hexapod walking robots are derived and validated through simulation which aid in the design and optimization of the robot parameters and workspace. To control the stability of the robot, a novel stability margin based on the normal contact forces of the robot is developed and then modified to account for the geometrical and physical attributes of the robot. The margin and its modified version are validated by comparison with a widely known stability criterion through simulated and physical experiments. A control scheme is developed to integrate the workspace and stability of multi-legged walking robots resulting in a bio-inspired reactive control strategy which is validated experimentally.

Robot swarms are envisioned in applications such as surveillance, agriculture, search-and-rescue operations, and construction. The decentralized nature of swarm intelligence has three key advantages over traditional multi-robot control algorithms: it is scalable, it is fault tolerant, and it is not susceptible to a single point of failure. These advantages are critical to the task of persistent surveillance - where a number of target locations need to be visited as frequently as possible. Unfortunately, in the real world, the autonomous robots that can be used for persistent surveillance have a limited battery life (or fuel capacity). Thus, they need to abandon their surveillance duties to visit a battery swapping station (or refueling depot) a.k.a. €˜depots€™. This €˜down time€™ reduces the frequency of visitation. This problem can be eliminated if the depots themselves were autonomous vehicles that could meet the (surveillance) robots at some point along their path from one target to another. Thus, the robots would spend less time on the 'charging' (or refueling) task. In this thesis we present decentralized control algorithms, and their results, for three stages of the persistent surveillance problem. First, we consider the case where the robots have no energy constraints, and use a decentralized approach to allow the robots choose the €˜best€™ target that they should visit next. While the selection process is decentralized, the robots can communicate with all the other robots in the swarm, and let them know which is their chosen target. We then consider the energy constraints of the robots, and slightly modify the algorithm, so that the robots visit a depot before they run out of energy. Lastly, we consider the case where the depots themselves can move, and communicate with the robots to pick a location and time to meet, to be able to swap the empty battery of a robot, with a fresh one. The goal of persistent surveillance is to visit target locations as frequently as possible, and thus, the performance measurement parameter is chosen to be the median frequency of visitation for all target locations. We evaluate the performance of the three algorithms in an extensive set of simulated experiments.

As the field of autonomous robots continue to advance, there is still a tremendous benefit to research human-supervised robot systems for fielding them in practical applications. The DRC inspired by the Fukushima nuclear power plant disaster has been a major research and development program for the past three years, to advance the field of human supervised control of robots for responding to natural and man-made disasters. The overall goal of the research presented in this thesis is to realise a new approach for semi-autonomous control of the Atlas humanoid robot under discrete commands from the human operator. A combination of autonomous and semi-autonomous perception and manipulation techniques to accomplish the task of detecting, opening and walking through a door are presented. The methods are validated in various different scenarios relevant to DRC door task.

The present work makes a study of the drag forces of umbilical cables of pipes inspection robots and aims to determine a valid model for their calculation. For this purpose, the physical models are developed in relation to the friction forces for the straight and curved sections in all their possible configurations, obtaining for each case a differential equation that models the friction phenomenon and that includes fluid and cable the characteristics. By solving the differential equation, an explicit expression is obtained for the value of the frictional force in each case. In the present work, the obtained model is evaluated in comparison with the traditional simplified model. It is also presented a study about the influence of the stiffness of the cable and how this characteristic generates the forces of the normal general forces at the ends of the curves that, finally, result in an additional component of the frictional forces. By comparison with actual drag force measurements, the model obtained is validated. The complexity of the calculations required for the determination of the pull forces in real pipes makes it necessary to use software tools to facilitate it. For this reason, two software applications are developed, one specific for rigid pipes and another one for flexible pipes. A series of strategies of how to keep low the maximum value of the cable tension are presented too. / El presente trabajo se dedica al estudio de fuerzas de arrastre de cables umbilicales de robot de inspección o desobstrucción de tuberías y tiene como objetivo la determinación de un modelo válido para el cálculo de las mismas. Para tal fin, se desarrollan los modelos físicos en relación a las fuerzas de rozamiento presentes para tramos rectos y curvas en todas sus configuraciones posibles, obteniéndose así, para cada caso, una ecuación diferencial que modela el fenómeno del rozamiento y que incluye las variables propias del fluido presente y las características pertinentes del cable. Mediante la resolución de esa ecuación diferencial, se obtiene una expresión explícita para el valor de la fuerza de rozamiento en cada caso. En el presente trabajo, el modelo obtenido es evaluado mediante comparación con el modelo simplificado tradicional de manera de ver los alcances del nuevo modelo. Se realiza también una presentación de la influencia que tiene la rigidez del cable y como esta rigidez a la flexión genera fuerzas normales adicionales en los extremos de las curvas que redundan finalmente en una componente adicional de las fuerzas de fricción. Mediante comparación con mediciones reales de fuerza de arrastre se valida el modelo obtenido Dada la complejidad de los cálculos requeridos para la determinación de las fuerzas de arrastre de tuberías reales se desarrollan dos aplicaciones de software para el cálculo de las mismas, una específica para tuberías rígidas y otra para tuberías flexibles utilizando algoritmos de cálculo basados en las ecuaciones determinadas. Son presentadas también, una serie de estrategias de forma de mantener acotado el valor máximo de la tensión de cable. Dentro de las estrategias están la elección de materiales de recubrimiento de cable que presenten un bajo coeficiente de fricción con la tubería, la adopción de cables cuyo peso específico sea tal que el peso se iguale a la fuerza de empuje y la incorporación de varios dispositivos de tracción del cable de modo de mantener bajas las contra tensiones del cable utilizando un sistema distribuido de empuje.

Lam Kin Kwan. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2005. / Includes bibliographical references (leaves 67-70). / Abstracts in English and Chinese. / Chapter 1. --- Introduction --- p.1 / Chapter 1.1 --- Visual Servoing --- p.1 / Chapter 1.1.1 --- System Architectures --- p.2 / Chapter 1.1.1.1 --- Position-based Visual Servoing --- p.2 / Chapter 1.1.1.2 --- Image-based Visual Servoing --- p.3 / Chapter 1.1.2 --- Camera Configurations --- p.4 / Chapter 1.2 --- Problem Definition --- p.5 / Chapter 1.3 --- Related Work --- p.6 / Chapter 1.4 --- Contribution of This Work --- p.9 / Chapter 1.5 --- Organization of This Thesis --- p.10 / Chapter 2. --- System Modeling --- p.11 / Chapter 2.1 --- Coordinate Frames --- p.11 / Chapter 2.2 --- System Kinematics --- p.13 / Chapter 2.3 --- System Dynamics --- p.14 / Chapter 2.4 --- Camera Model --- p.15 / Chapter 2.4.1 --- Eye-in-hand Configuration --- p.18 / Chapter 2.4.2 --- Eye-and-hand Configuration --- p.21 / Chapter 3. --- Adaptive Visual Servoing Control --- p.24 / Chapter 3.1 --- Controller Design --- p.24 / Chapter 3.2 --- Parameter Estimation --- p.27 / Chapter 3.3 --- Stability Analysis --- p.30 / Chapter 4. --- Experimental Studies --- p.34 / Chapter 4.1 --- Experimental Setup --- p.34 / Chapter 4.1.1 --- Hardware Setup --- p.34 / Chapter 4.1.2 --- Image Pattern Recognition --- p.35 / Chapter 4.1.3 --- Experimental Task --- p.36 / Chapter 4.2 --- Control Performance with Different Proportional Gains and Derivative Gains --- p.39 / Chapter 4.3 --- Control Performance with Different Adaptive Gains --- p.41 / Chapter 4.4 --- Gravity Compensator --- p.50 / Chapter 4.5 --- Control Performance with Previous Image Positions --- p.51 / Chapter 4.6 --- Kinematic Controller --- p.56 / Chapter 5. --- Conclusions --- p.61 / Chapter 5.1 --- Conclusions --- p.61 / Chapter 5.2 --- Future Work --- p.62 / Appendix --- p.63 / Bibliography --- p.67

Within the field of multi-robot control, there is a large focus in research involving consensus. In this thesis two parts will be studied. The first development of this thesis is a consensus-based robot arm platform. To implement, two robotic arms are developed and studied. The most effective robot arm is then utilized to create a robot arm network testbed. Consensus is used to coordinate several robot arms and decentralize system computation. The research explores a platform to facilitate consensus on a group of robotic arms. The second development is in Cartesian coordinate collective motion. This collective motion control combines consensus through coupling of Cartesian coordinates. The controller is presented with simulation and experimental validation. Integration of both parts of the thesis is then discussed in application. An example is provided to demonstrate usefulness. In conclusion, this thesis provides more control to a system of ground robots using collective motion and consensus-based robot arms.

Giving robots the ability to move around autonomously in various real-world environments has long been a major challenge for Artificial Intelligence. New approaches to the design and control of autonomous robots have shown the value of drawing inspiration from the natural world. Animals navigate, perceive and interact with various uncontrolled environments with seemingly little effort. Flying insects, in particular, are quite adept at manoeuvring in complex, unpredictable and possibly hostile environments.Inspired by the miniature machine view of insects, this thesis contributes to the autonomous control of mobile robots through the application of insect-based visual cues and behaviours. The parsimonious, yet robust, solutions offered by insects are directly applicable to the computationally restrictive world of autonomous mobile robots. To this end, two main navigational domains are focussed on: corridor guidance and visual homing.Within a corridor environment, safe navigation is achieved through the application of simple and intuitive behaviours observed in insect, visual navigation. By observing and responding to observed apparent motions in a reactive, yet intelligent way, the robot is able to exhibit useful corridor guidance behaviours at modest expense. Through a combination of both simulation and real-world robot experiments, the feasibility of equipping a mobile robot with the ability to safely navigate in various environments, is demonstrated.It is further shown that the reactive nature of the robot can be augmented to incorporate a map building method that allows previously encountered corridors to be recognised, through the observation of landmarks en route. This allows for a more globally-directed navigational goal.Many animals, including insects such as bees and ants, successfully engage in visual homing. This is achieved through the association of ++ / visual landmarks with a specific location. In this way, the insect is able to 'home in' on a previously visited site by simply moving in such a way as to maximise the match between the currently observed environment and the memorised 'snapshot' of the panorama as seen from the goal. A mobile robot can exploit the very same strategy to simply and reliably return to a previously visited location.This thesis describes a system that allows a mobile robot to home successfully. Specifically, a simple, yet robust, homing scheme that relies only upon the observation of the bearings of visible landmarks, is proposed. It is also shown that this strategy can easily be extended to incorporate other visual cues which may improve overall performance.The homing algorithm described, allows a mobile robot to home incrementally by moving in such a way as to gradually reduce the discrepancy between the current view and the view obtained from the home position. Both simulation and mobile robot experiments are again used to demonstrate the feasibility of the approach.