With an increased push for commercial autonomous cars, the demand of high speed systems capable of performing in unstructured driving environments is growing. In this thesis, the behavior of a bio-inspired predator prey model is considered to stimulate a more organic response to obstacles and a moving target than existing algorithms. However, the current predator prey model has a disconnect between the desired velocities commanded and the torque signals provided to the motors due the dynamics of the vehicle not accounted for. This causes the vehicle to derail from its intended trajectory at sharp turns. In this study, we start by adding dynamic behavior to the unicycle model to account for the varying dynamics of the vehicle. A backstepping algorithm is developed to connect the predator-prey model commanding desired velocities to an appropriate torque controller for the motors of the vehicle. To account for the unknown dynamic model parameters an adaptive control approach is utilized. Three different controllers are developed and evaluated. Out of the three, the indirect MRAC backstepping controller is deemed unsuitable due to its limitations with handling unknown parameter structure. The direct MRAC backstepping is deemed suitable and therefore simulated and implemented on the vehicle. The newly derived controller is able to overcome the disconnect and allow the vehicle to optimally track its trajectory for a velocity range of 1 m/s to 9 m/s despite varying dynamics. Lastly, the L1 adaptive backstepping controller is introduced and simulated to provide an alternative, more robust solution to the direct MRAC backstepping controller. / Master of Science

A differential steered, 13.6 kg robot was developed as an intelligent tire testing system and was used to investigate the potential of using piezoelectric film sensors in small tube-type pneumatic tires to characterize tire-ground interaction.<br />One focus of recent research in the tire industry has been on instrumenting tires with sensors to monitor the tire, vehicle, or external environment. On small robots, tire sensors that measure the forces and deflections in the contact patch could be used to improve energy efficiency and/or mobility during a mission.<br />The robot was assembled from a SuperDroid Robots kit and instrumented with low-cost piezoelectric film sensors from Measurement Specialties between the inner tube and the tire.  An unlaminated and a laminated sensor were placed circumferentially along the tread and an unlaminated sensor was placed along the sidewall.  A slip ring transferred the signals from the tire to the robot. There, the signal conditioning circuit extended the time constant of the sensors and filtered electromagnetic interference.  The robot was tested with a controlled power sequence carried out on polished cement, ice, and sand at three power levels, two payload levels, and with two tire sizes.<br />The results suggest that the sensors were capable of detecting normal pressure, deflection, and/or longitudinal strain.  Added payload increased the amplitude of the signals for all sensors.  On the smaller tires, sensors generally recorded a smaller, wider signal on sand compared to cement, indicating the potential to detect contact patch pressure and length.  The signals recorded by the unlaminated sensor along the tread of the smaller tire were lower on ice compared to cement, indicating possible sensitivity to tractive force.  Results were less consistent for the larger tires, possibly due to the large tread pattern. / Master of Science

The work presented in this thesis focuses on the design and testing of a casualty extraction robot intended to stabilize the head and neck of an unresponsive person. The employment of robots in dangerous locales such as combat zones or the site of a natural disaster has the potential to help keep first responders out of harm's way as well as to improve the efficiency of search and rescue teams. After a review of robotic search and rescue platforms the Semi-Autonomous Victim Extraction Robot(SAVER) is introduced. The necessity of a device intended to support the head and cervical spine during transport on a rescue robot is then discussed. The kinematic and dynamic analyses of various candidate differential mechanisms intended for the head stabilization device are described, and the chosen mechanism is demonstrated in a proof-of-concept device. Following testing with a simple PID controller, it was determined an advanced feedback controller with disturbance rejection capabilities was required. Linear Active Disturbance Rejection Control (LADRC) was chosen for its effectiveness in rejecting perturbations and handling modeling uncertainties. The performance the proposed LADRC control scheme was compared with PID in simulation and the results are presented. Finally, a prototype of the device was designed and built to validate the functionality of the subsystem, and the results of the corresponding experimentation are discussed. / M. S. / Robots can help to keep first responders and medics out of dangerous situations by performing the rescue operation themselves or by collaborating with the field medic to make the process quicker and more efficient. The work presented in this thesis begins with a review of state-of-the-art rescue robots followed by the a brief description of the design of a Semi-Autonomous Victim Extraction Robot (SAVER) intended to rescue injured and incapacitated people. After the SAVER system is briefly described, the necessity of a device intended to support the head and cervical spine during transport is discussed. The head stabilization subsystem could also be implemented as a standalone device for use by paramedics to help free up valuable time that would otherwise be spent in manually stabilizing the head and neck of the injured person

This thesis explores the field of legged robots with reduced degree-of-freedom (DOF) leg mechanisms. Multi-legged robots have drawn interest among researchers due to their high level of adaptability on unstructured terrains. However, conventional legged robots require multiple degrees of freedom and each additional degree of freedom increases the overall weight and complexity of the system. Additionally, the complexity of the control algorithms must be increased to provide mobility, stabilization, and maneuvering. Normally, robotic legs are designed with at least three degrees of freedom resulting in complex articulated mechanisms, which limits the applicability of such robots in real-world applications. However, reduced DOF leg mechanisms come with reduced tasking capabilities, such as maintaining constant body height and velocity during locomotion. To address some of the challenges, this thesis proposes a novel bipedal robot with reduced DOF leg mechanisms. The proposed leg mechanism utilizes the Reuleaux triangle to generate the foot trajectory to achieve a constant body height during locomotion while maintaining a constant velocity. By using a differential drive, the robot is also capable of steering. In addition to the analytical results of the trajectory profile of each leg, the thesis provides a trajectory function of the Reuelaux triangle cam with respect to time such that the robot can maintain a constant velocity and constant body height during walking. An experimental prototype of the bipedal robot was integrated and experiments were conducted to evaluate the walking capability of the robot. Ongoing future work of the proposed design is also outlined in the thesis. / Master of Science / Bipedal robots are a type of legged robots that use two legs to move. Legs require multiple degrees of freedom to provide propulsion, stabilization, and maneuvering. Additional degrees of freedom of the leg result in a heavier robot, more complex control method, and more energy consumption. However, reduced degree of freedom legs result in a tradeoff between certain tasking capabilities for easier controls and lower energy consumption. As an attempt to overcome these challenges, this thesis presents a robot design with a reduced degree of freedom leg mechanism. The design of the mechanism is described in detail with its preliminary analysis. In addition, this thesis presents experimental validation with the robot which validates that the robot is capable of moving with constant body height at constant velocity while being of capable of steering. The thesis concludes with a discussion of the future work.

Instability of humanoid robots is a common problem, especially given external disturbances or difficult terrain. Even with the robustness of most whole body controllers, instability is inevitable given the right conditions. When these unstable events occur they can result in costly damage to the robot potentially causing a cease of normal functionality. Therefore, it is important to study and develop methods to control a humanoid robot during a fall to reduce the chance of critical damage. This thesis proposes joint angular velocity strategies to reduce the impact velocity resulting from a lateral, backward, or forward fall. These strategies were used on two and three link reduced order models to simulate a fall from standing height of a humanoid robot. The results of these simulations were then used on a full degree of freedom robot, Viginia Tech's humanoid robot ESCHER, to validate the efficacy of these strategies. By using angular velocity strategies for the knee and waist joint, the reduced order models resulted in a decrease in impact velocity of the center of mass by 58%, 87%, and 74% for a lateral, backward, and forward fall respectively in comparison to a rigid fall using the same initial conditions. Best case angular velocity strategies were then developed for various initial conditions for each falling direction. Finally, these parameters were implemented on the full degree of freedom robot which showed results similar to those of the reduced order models. / Master of Science / Instability of humanoid robots is a common problem, especially given external disturbances or difficult terrain. Even with the robustness of most whole body controllers, instability is inevitable given the right conditions. When these unstable events occur they can result in costly damage to the robot potentially causing a cease of normal functionality. Therefore, it is important to study and develop methods to control a humanoid robot during a fall to reduce the chance of critical damage. This thesis proposes strategies that rotate the joints at a constant rate to reduce damage resulting from a lateral, backward, or forward fall. These strategies were used on two and three link simplistic models to simulate a fall from standing height of a humanoid robot. The results of these simulations were then used on a full robot, Viginia Tech’s humanoid robot ESCHER, to validate the efficacy of these strategies. By constant joint rotation strategies for the knee and waist joint, the simplistic models resulted in a decrease in impact velocity of the center of mass by 58%, 87%, and 74% for a lateral, backward, and forward fall respectively in comparison to a rigid fall using the same initial conditions. Best case joint rotation strategies were then developed for various initial conditions for each falling direction. Finally, these parameters were implemented on the full robot which showed results similar to those of the reduced order models.

This thesis presents the design and control of a cable-actuated mobile snake robot. The goal of this research is to reduce the size of snake robots and improve their locomotive efficiency by simultaneously actuating groups of links to fit optimized curvature profiles. The basic functional unit of the snake is a four-link, single degree of freedom module that bends using an antagonistic cable-routing scheme. Elastic elements in series with the cables and the coupled nature of the mechanism allow each module to detect and automatically respond to obstacles. The mechanical and electrical designs of the bending module are presented, with emphasis on the cable-routing scheme, key optimizations, and the use of series elastic actuation. An approximate expression for the propulsive force generated by a snake as a function of its articulation (i.e. the number of links it contains divided by its body length) is derived and a closed-form approximation for the optimal phase offset between joints to maximize the speed of a snake is obtained by simplifying a previous result. A simplified model of serpentine locomotion that considers the forces acting on a single link as it traverses a sinusoid is presented and compared to a detailed multibody dynamic model. Control strategies for snake robots with coupled joints are developed, along with a feedback linearization of the joint dynamics. Experimental studies of force control, locomotion, and adaptation to obstacles using a fully integrated prototype are presented and compared with simulated results. / MS / This thesis presents the development of a cable-driven snake robot, with the goal of decreasing the size and mass of these devices and increasing their efficiency. Snake robots have potential applications in exploration and manipulation in cluttered or confined environments. The cable transmission system presented in this thesis allows for multiple links in a snake robot to be actuated simultaneously, allowing for increased articulation in a robot of fixed size and mass. Serpentine locomotion, in which a sinusoidal wave is propagated down the robot’s length, is a silent and energy-efficient mode of transportation, widely employed in the animal kingdom. Snake robots achieve serpentine locomotion by driving their joints sinusoidally, with adjacent joints moving asynchronously, with the time lag between joints set by the value of a phase offset. An expression for the optimal phase offset to maximize forward velocity is derived by simplifying a previous result from the literature. An approximation of the dynamics of serpentine locomotion for a snake traveling at constant velocity is then derived, and this model is used to obtain an approximate limiting expression for the propulsive force generated per link as a function of the number of links in the snake. Methods to control a snake composed of coupled linkages are explored and the mechatronic design of a fully integrated prototype is presented. Experiments on force control, locomotion and turning, and detection and interaction with obstacles using the prototype are then described.

L'architecture mécanique, l'actionnement, la détection directe ou indirecte des efforts ainsi que la conception de contrôleurs en impédance ou en admittance sont les aspects fondamentaux et importants à considérer pour le développement d'un robot permettant une interaction physique humain-robot (IPHR) sécuritaire. Cette thèse est consacrée au développement de nouvelles architectures de robots pour l'IPHR qui ont une structure simple, peu ou pas de singularités, qui sont légers et à faible impédance mécanique. Un nouveau robot parallèle hybride cinématiquement redondant (RPHCR) sans singularité dans l'espace de travail et ayant une faible inertie mobile est d'abord proposé. Le concept de la redondance cinématique des membrures et l'agencement d'assemblage de la plate-forme mobile de ce robot sont ensuite généralisés et développés en une méthodologie pour la synthèse de nouveaux RPHCRs. Plusieurs exemples d'architectures sont présentés et une solution analytique du problème géométrique inverse est obtenue. Le problème géométrique direct des RPHCRs doit être résolu afin de déterminer la position et l'orientation de la plate-forme mobile pour des coordonnées articulaires données. Différentes approches pour résoudre le problème géométrique direct sont alors proposées. Il est montré que le problème géométrique direct des RPHCRs proposés dans la thèse est beaucoup plus simple que celui associé aux robots non redondants ou à de nombreux autres robots parallèles cinématiquement redondants. L'agrandissement de l'espace de travail et l'optimisation des trajectoires articulaires des RPHCRs sont réalisés en déterminant les valeurs optimales des coordonnées redondantes. Enfin, la redondance est en outre utilisée pour opérer un préhenseur monté sur la plateforme mobile à partir des actionneurs fixés à la base du robot ou près de celle-ci. Un contrôleur combiné en position et force de préhension est proposé pour le contrôle de la force de préhension. / Robot architecture, actuation, indirect/direct force sensing, and impedance/admittance controller design are the fundamental and important aspects to be considered in order to achieve safe physical human-robot interaction (pHRI). This thesis is devoted to the development of novel robot architectures for pHRI that have a simple structure, few or no singularities, lightweight, and low-impedance. A novel kinematically redundant hybrid parallel robot (KRHPR) that is singularity-free throughout the workspace and has low moving inertia is firstly proposed. The concept of the redundant links and moving platform assembly arrangement of this robot is further generalised and developed into a methodology for the synthesis of novel KRHPRs. Several example architectures are presented and an analytical inverse kinematic solution is derived.The forward kinematics of the KRHPRs must be solved to determine the position and orientation of the moving platform for given joint coordinates. Different approaches for solving the forward kinematic problem are then proposed. It is shown that the forward kinematics of the KRHPRs proposed in the thesis is much simpler than that of their non-redundant counterparts or that of many other kinematically redundant parallel robots. Workspace enlargement and joint trajectory optimisation of the KRHPRs are pursued by determining the optimal values of the redundant coordinates. Finally, the redundancy is further utilised to operate a gripper on the moving platform from the base actuators. A combined position and grasping force controller is proposed for the control of the grasping force.

Finding a fire inside of a structure without knowing its conditions poses a dangerous threat to the safety of firefighters. As a result, robots are being explored to increase awareness of the conditions inside structures before having firefighter enter. This thesis presents a method that autonomously guides a robot to the location of a fire inside a structure. The method uses classification of fire, smoke, and other fire environment objects to calculate a weighted probability. Weighted probability is a measurement that indicates the probability that a given region on an infra-red image will lead to fire. This method was tested on large-scale fire videos with a robot moving towards a fire and it is also compared to following the highest temperatures on the image. Sending a robot to find a fire has the potential to save the lives of firefighters. / Master of Science / Finding a fire inside of a structure without knowing its conditions poses a dangerous threat to the safety of firefighters. As a result, robots are being explored to increase awareness of the conditions inside structures before having firefighter enter. This thesis presents a method that autonomously guides a robot to the location of a fire inside a structure. The method uses classification of fire, smoke, and other fire environment objects to calculate a weighted probability. Weighted probability is a measurement that indicates the probability that a given region on an infra-red image will lead to fire. This method was tested on large-scale fire videos with a robot moving towards a fire and it is also compared to following the highest temperatures on the image. Sending a robot to find a fire has the potential to save the lives of firefighters.

Each year, falls from elevated structures, like scaffolding, kill or seriously injure over a thousand construction workers (Bureau of Labor Statistics, 2007). To prevent such falls, the development of a robotic system is proposed that can climb and navigate on the complex structures, performing hazardous inspection and maintenance in place of humans. In this work, a serpentine robotic system is developed that will be able to climb pole-like structures, such as scaffolding and trusses, commonly found on work sites. Serpentine robots have been proven to be effective at traversing unstructured terrains and manipulating complex objects. The work presented in this thesis adds a new method of mobility for serpentine robots, specifically those with actuated universal joint structures. Movement is produced by inducing a wobbling motion between adjacent modules through oscillatory motions in the actuated axis of the universal joint. Through the frictional interactions between the modules of the serpentine and the surface of the pole, the wobbling motion lets the serpentine effectively roll up the pole's surface. This work investigates theoretical and experimental results for a serpentine robot climbing a pole structure. It discusses the structure and design parameters of the robot and develops relationships between them. These geometric and performance-based relationships are then used to create a design space that provides a guide for choosing a combination of module dimensions for a desired set of performance parameters. From this, case studies are shown which give examples of how the design space can be used for several different applications. Based on the design space procedure, a serpentine robot, HyDRAS (Hyper-Redundant Discrete Robotic Articulated Serpentine) was designed and built. The robot was used to prove the validity of the design space procedure and to validate the climbing motion algorithms. Several tests were performed with HyDRAS that showed the practicality of the helical rolling motion, as well as the feasibility of serpentine pole climbing. Observations and discussion based on the experiments are given, along with the plans for future work involving pole-climbing serpentine robots. / Master of Science

Depuis plusieurs années la robotique est vue comme une solution clef pour améliorer la qualité de vie des personnes ayant subi une amputation. Pour créer de nouvelles prothèses intelligentes qui peuvent être facilement intégrées à la vie quotidienne et acceptée par ces personnes, celles-ci doivent être non-intrusives, fiables et peu coûteuses. L’électromyographie de surface fournit une interface intuitive et non intrusive basée sur l’activité musculaire de l’utilisateur permettant d’interagir avec des robots. Cependant, malgré des recherches approfondies dans le domaine de la classification des signaux sEMG, les classificateurs actuels manquent toujours de fiabilité, car ils ne sont pas robustes face au bruit à court terme (par exemple, petit déplacement des électrodes, fatigue musculaire) ou à long terme (par exemple, changement de la masse musculaire et des tissus adipeux) et requiert donc de recalibrer le classifieur de façon périodique. L’objectif de mon projet de recherche est de proposer une interface myoélectrique humain-robot basé sur des algorithmes d’apprentissage par transfert et d’adaptation de domaine afin d’augmenter la fiabilité du système à long-terme, tout en minimisant l’intrusivité (au niveau du temps de préparation) de ce genre de système. L’aspect non intrusif est obtenu en utilisant un bracelet à électrode sèche possédant dix canaux. Ce bracelet (3DC Armband) est de notre (Docteur Gabriel Gagnon-Turcotte, mes co-directeurs et moi-même) conception et a été réalisé durant mon doctorat. À l’heure d’écrire ces lignes, le 3DC Armband est le bracelet sans fil pour l’enregistrement de signaux sEMG le plus performant disponible. Contrairement aux dispositifs utilisant des électrodes à base de gel qui nécessitent un rasage de l’avant-bras, un nettoyage de la zone de placement et l’application d’un gel conducteur avant l’utilisation, le brassard du 3DC peut simplement être placé sur l’avant-bras sans aucune préparation. Cependant, cette facilité d’utilisation entraîne une diminution de la qualité de l’information du signal. Cette diminution provient du fait que les électrodes sèches obtiennent un signal plus bruité que celle à base de gel. En outre, des méthodes invasives peuvent réduire les déplacements d’électrodes lors de l’utilisation, contrairement au brassard. Pour remédier à cette dégradation de l’information, le projet de recherche s’appuiera sur l’apprentissage profond, et plus précisément sur les réseaux convolutionels. Le projet de recherche a été divisé en trois phases. La première porte sur la conception d’un classifieur permettant la reconnaissance de gestes de la main en temps réel. La deuxième porte sur l’implémentation d’un algorithme d’apprentissage par transfert afin de pouvoir profiter des données provenant d’autres personnes, permettant ainsi d’améliorer la classification des mouvements de la main pour un nouvel individu tout en diminuant le temps de préparation nécessaire pour utiliser le système. La troisième phase consiste en l’élaboration et l’implémentation des algorithmes d’adaptation de domaine et d’apprentissage faiblement supervisé afin de créer un classifieur qui soit robuste au changement à long terme. / For several years, robotics has been seen as a key solution to improve the quality of life of people living with upper-limb disabilities. To create new, smart prostheses that can easily be integrated into everyday life, they must be non-intrusive, reliable and inexpensive. Surface electromyography provides an intuitive interface based on a user’s muscle activity to interact with robots. However, despite extensive research in the field of sEMG signal classification, current classifiers still lack reliability due to their lack of robustness to short-term (e.g. small electrode displacement, muscle fatigue) or long-term (e.g. change in muscle mass and adipose tissue) noise. In practice, this mean that to be useful, classifier needs to be periodically re-calibrated, a time consuming process. The goal of my research project is to proposes a human-robot myoelectric interface based on transfer learning and domain adaptation algorithms to increase the reliability of the system in the long term, while at the same time reducing the intrusiveness (in terms of hardware and preparation time) of this kind of systems. The non-intrusive aspect is achieved from a dry-electrode armband featuring ten channels. This armband, named the 3DC Armband is from our (Dr. Gabriel Gagnon-Turcotte, my co-directors and myself) conception and was realized during my doctorate. At the time of writing, the 3DC Armband offers the best performance for currently available dry-electrodes, surface electromyographic armbands. Unlike gel-based electrodes which require intrusive skin preparation (i.e. shaving, cleaning the skin and applying conductive gel), the 3DC Armband can simply be placed on the forearm without any preparation. However, this ease of use results in a decrease in the quality of information. This decrease is due to the fact that the signal recorded by dry electrodes is inherently noisier than gel-based ones. In addition, other systems use invasive methods (intramuscular electromyography) to capture a cleaner signal and reduce the source of noises (e.g. electrode shift). To remedy this degradation of information resulting from the non-intrusiveness of the armband, this research project will rely on deep learning, and more specifically on convolutional networks. The research project was divided into three phases. The first is the design of a classifier allowing the recognition of hand gestures in real-time. The second is the implementation of a transfer learning algorithm to take advantage of the data recorded across multiple users, thereby improving the system’s accuracy, while decreasing the time required to use the system. The third phase is the development and implementation of a domain adaptation and self-supervised learning to enhance the classifier’s robustness to long-term changes.