Robots are increasingly becoming safer and more capable. In the past, the main applications for robots have been in manufacturing, where they perform repetitive, highly accurate tasks with physical barriers that separate them from people. They have also been used in space exploration where people are not around. Due to improvements in sensors, algorithms, and design, robots are beginning to be used in other applications like materials handling, healthcare, and agriculture and will one day be ubiquitous. For this to be possible, they will need to be able to function safely in unmodelled and dynamic environments. This is especially true when working in a shared space with people. We desire for robots to interact with people in a way that is helpful and intuitive. This requires that the robots both act predictably and be able to predict short-term human intent. We create a model for predicting short-term human intent in a collaborative furniture carrying task that a robot could use to be a more responsive and intuitive teammate. For robots to perform collaborative manipulation tasks with people naturally and efficiently, understanding and predicting human intent is necessary. We completed an exploratory study recording motion and force for 21 human dyads moving an object in tandem in a variety of tasks to better understand how they move and how their movement can be predicted. Using the previous 0.75 seconds of data, the human intent can be predicted for the next 0.25 seconds. This can then be used with a robot in real applications. We also show that force data is not required to predict human intent. We show how the prediction data works in real-time, demonstrating that past motion alone can be used to predict short-term human intent. We show this with human-human dyads and a human-robot dyad. Finally, we imagine that soft robots will be common in human-robot interaction. We present work on controlling soft, pneumatically-actuated, inflatable robots. These soft robots have less inertia than traditional robots but a high power density which allows them to operate in proximity to people. They can, however, be difficult to control. We developed a neural net model to use for control of our soft robot. We have shown that we can predict human intent in a human-robot dyad which is an important goal in physical human-robot interaction and will allow robots to co-manipulate objects with humans in an intelligent way.

As robots become more common operating in close proximity to people, new opportunities arise for physical human-robot interaction, such as co-manipulation of extended objects. Co-manipulation involves physical interaction between two partners where an object held by both is manipulated in tandem. There is a dearth of viable high degree-of-freedom co-manipulation controllers, especially for extended objects, as well as a lack of information about how human-human teams perform in high degree-of-freedom tasks. One method for creating co-manipulation controllers is to pattern them off of human data. This thesis uses this technique by exploring a previously completed experimental study. The study involved human-human dyads in leader-follower format performing co-manipulation tasks with an extended object in 6 degrees of freedom. Two important tasks performed in this experiment were lateral translation and planar rotation tasks. This thesis focuses on these two tasks because they represent planar motion. Most previous control methods are for 1 or 2 degrees-of-freedom. The study provided information about how human-human dyads perform planar tasks. Most notably, planar tasks generally adhere to minimum-jerk trajectories, and do not minimize interaction forces between users. The study also helped solve the translation versus rotation problem. From the experimental data, torque patterns were discovered at the beginning of the trial that defined intent to translate or rotate. From these patterns, a new method of planar co-manipulation control was developed, called Extended Variable Impedance Control. This is a novel 3 degree-of-freedom method that is applicable to a variety of planar co-manipulation scenarios. Additionally, the data was fed through a Recursive Neural Network. The network takes in a series of motion data and predicts the next step in the series. The predicted data was used as an intent estimate in another novel 3 degree of freedom method called Neural Network Prediction Control. This method is capable of generalizing to 6 degrees of freedom, but is limited in this thesis for comparison with the other method. An experiment, involving 16 participants, was developed to test the capabilities of both controllers for planar tasks. A dual manipulator robot with an omnidirectional base was used in the experiment. The results from the study show that both the Neural Network Prediction Control and Extended Variable Impedance Control controllers performed comparably to blindfolded human-human dyads. A survey given to participants informed us they preferred to use the Extended Variable Impedance Control. These two unique controllers are the major results of this work.

Inspiration from biological systems suggests that robots should demonstrate same level of capabilities that are embedded in biological systems in performing safe and successful interaction with the humans. The major challenge in physical human robot interaction tasks in anthropic environment is the safe sharing of robot work space such that robot will not cause harm or injury to the human under any operating condition. Embedding human like adaptable compliance characteristics into robot manipulators can provide safe physical human robot interaction in constrained motion tasks. In robotics, this property can be achieved by using active, passive and semi active compliant actuation devices. Traditional methods of active and passive compliance lead to complex control systems and complex mechanical design. In this thesis we present compliant robot manipulator system with semi active compliant device having magneto rheological fluid based actuation mechanism. Human like adaptable compliance is achieved by controlling the properties of the magneto rheological fluid inside joint actuator. This method offers high operational accuracy, intrinsic safety and high absorption to impacts. Safety is assured by mechanism design rather than by conventional approach based on advance control. Control schemes for implementing adaptable compliance are implemented in parallel with the robot motion control that brings much simple interaction control strategy compared to other methods. Here we address two main issues: human robot collision safety and robot motion performance.We present existing human robot collision safety standards and evaluate the proposed actuation mechanism on the basis of static and dynamic collision tests. Static collision safety analysis is based on Yamada’s safety criterion and the adaptable compliance control scheme keeps the robot in the safe region of operation. For the dynamic collision safety analysis, Yamada’s impact force criterion and head injury criterion are employed. Experimental results validate the effectiveness of our solution. In addition, the results with head injury criterion showed the need to investigate human bio-mechanics in more details in order to acquire adequate knowledge for estimating the injury severity index for robots interacting with humans. We analyzed the robot motion performance in several physical human robot interaction tasks. Three interaction scenarios are studied to simulate human robot physical contact in direct and inadvertent contact situations. Respective control disciplines for the joint actuators are designed and implemented with much simplified adaptable compliance control scheme. The series of experimental tests in direct and inadvertent contact situations validate our solution of implementing human like adaptable compliance during robot motion and prove the safe interaction with humans in anthropic domains.

In the last decades research and development in the field of robotics has grown rapidly. This growth has resulted in the emergence of service robots that need to be able to physically interact with humans for different applications. One of these applications involves robots and humans cooperating in handling an object together. In such cases, there is usually an initial arrangement of how the robot and the humans hold the object and the arrangement stays the same throughout the manipulation task. Real-world scenarios often require that the initial arrangement changes throughout the task, therefore, it is important that the robot is able to recognize these changes and act accordingly. We consider a setting where a robot holds a large flat object with one or two humans. The aim of this research project is to detect the change in the number of agents grasping the object using only force and torque information measured at the robot's wrist. The proposed solution involves defining a transition sequence of four steps that the humans should perform to go from the initial scenario to the final one. The force and torque information is used to estimate the grasping point of the agents with a Kalman filter. While the humans are going from one scenario to the other, the estimated point changes according to the step of the transition the humans are in. These changes are used to track the steps in the sequence using a hidden Markov model (HMM). Tracking the steps in the sequence means knowing how many agents are grasping the object. To evaluate the method, humans that were not involved in the training of the HMM were asked to perform two tasks: a) perform the previously defined sequence as is, and b) perform a deviation of the sequence. The results of the method show that it is possible to detect the change between one human and two humans holding the object using only force and torque information.

Soft robots are inherently safer than traditional robots due to their compliance and high power density ratio resulting in lower accidental impact forces. Thus they are a natural option for human-robot interaction. This thesis specifically looked at human-robot co-manipulation which is defined as a human and a robot working together to move an object too large or awkward to be safely maneuvered by a single agent. To better understand how humans communicate while co-manipulating an object, this work looked at haptic interaction of human-human dyadic co-manipulation trials and studied some of the trends found in that interaction. These trends point to ways robots can effectively work with human partners in the future. Before successful human-robot co-manipulation with large-scale soft robots can be achieved, low-level joint angle control is needed. Low-level model predictive control of soft robot joints requires a sufficiently accurate model of the system. This thesis introduces a recursive Newton-Euler method for deriving the dynamics that is sufficiently accurate and accounts for flexible joints in an intuitive way. This model has been shown to be accurate to a median absolute error of 3.15 degrees for a three-link three-joint six degree of freedom soft robot arm. Once a sufficiently accurate model was developed, a gain-based evolutionary model predictive control (MPC) technique was formulated based on a previous evolutionary MPC technique. This new method is referred to as model evolutionary gain-based predictive control or MEGa-PC. This control law is compared to nonlinear evolutionary model predictive control (NEMPC). The new technique allows intentionally decreasing the control frequency to 10 Hz while maintaining control of the system. This is proven to help MPC solve more difficult problems by having the ability to extend the control horizon. This new controller is also demonstrated to work well on a three-joint three-link soft robot arm. Although complete physical human-robot co-manipulation is outside the scope of this thesis, this thesis covers three main building blocks for physical human and soft robot co-manipulation: human-human haptic communication, soft robot modeling, and model evolutionary gain-based predictive control.

Joint torque feedback (JTF) is a new and promising means of kinesthetic feedback to provide information to a person or guide them during a motion task. However, little work has been done to apply the torque feedback to a person. This project evaluates the properties of JTF as haptic feedback, starting from the fabrication of a lightweight elbow haptic exoskeleton. A cheap hobby motor and easily accessible hardware are introduced for manufacturing and open-sourced embedded architecture for data logging. The total cost and the weights are $500 and 509g. Also, as the prerequisite step to assess the JTF in guidance, human perceptual ability to detect JTF was quantified at the elbow during all possible static and dynamic joint statuses. JTF slopes per various joint conditions are derived using the Interweaving Staircase Method. For either directional torque feedback, flexional motion requires 1.89-2.27 times larger speed slope, in mNm/(°/s), than the extensional motion. In addition, we find that JTFs during the same directional muscle's isometric contraction yields a larger slope, in mNm/mNm, than the opposing direction (7.36 times and 1.02 times for extension torque and flexion torque). Finally, the guidance performance of the JTF was evaluated in terms of time delay and position error between the directed input and the wearer's arm. When studying how much the human arm travels with JTF, the absolute magnitude of the input shows more significance than the duration of the input (p-values of <0.0001 and 0.001). In the analysis of tracking the pulse input, the highest torque stiffness, 95 mNm/°, is responsible for the smallest position error, 6.102 ± 5.117°, despite the applied torque acting as compulsory stimuli. / Doctor of Philosophy / Joint torque feedback (JTF) is a new and promising means of haptic feedback to provide information to a person or guide them during a motion task. However, little work has been done to apply the torque feedback to a person, such as determining how well humans can detect external torques or how stiff the torque input should be to augment a human motion without interference with the voluntary movement. This project evaluates the properties of JTF as haptic feedback, starting from the fabrication of a lightweight elbow haptic exoskeleton. The novelty of the hardware is that we mask most of the skin receptors so that the joint receptors are primarily what the body will use to detect external sensations. A cheap hobby motor and easily accessible hardware are introduced for manufacturing and open-sourced software architecture for data logging. The total cost and the weight are $500 and 509g. Also, as the prerequisite step to assess the JTF in guidance, human perceptual ability to detect JTF was quantified at the elbow during all possible static and dynamic joint statuses. A psychophysics tool called Interweaving Staircase Method was implemented to derive torque slopes per various joint conditions. For either directional torque feedback, flexional motion requires 1.89-2.27 times larger speed slope, in mNm/(°/s) than the extensional motion. In addition, the muscles' isometric contraction with the aiding direction required a larger slope, in $mathrm{mNm/mNm}$ than the opposing direction (7.36 times and 1.02 times for extension torque and flexion torque). Finally, the guidance performance of the JTF was evaluated in terms of time delay and position error between the directed input and the wearer's arm. When studying how much the human arm travels with JTF, the absolute magnitude of the input shows more significance than the duration of the input (p-values of <0.0001 and 0.001). In the analysis of tracking the pulse input, the highest torque stiffness, 95 mNm/°, is responsible for the smallest position error, 6.102 ± 5.117°, despite the applied torque acting as compulsory stimuli.

Au cours des dernières années, nous avons assisté à un changement de paradigme, passant de la fabrication de robots rigides à des robots compliants. Ceci est dû à plusieurs raisons telles que l'amélioration de l'efficacité des robots dans la réalisation des mouvements explosifs ou cycliques. En fait, l'une des premières motivations à l'origine de ce changement est la sécurité. Parlant de la sécurité à la fois du sujet humain et du robot, tout en s'engageant dans des tâches collaboratives. Ainsi la désignation des cobots. Les cobots peuvent aider un opérateur humain expérimenté dans plusieurs domaines où la précision est essentielle, comme les applications industrielles ou les tâches médicales. Jusqu'à présent, les cobots présentent toujours des problèmes de sécurité, même avec des recommandations réglementaires telles que ISO / TS 15066 et ISO 10218-1 et 2 qui limitent leurs avantages économiques. Dans cette vue, plusieurs projets de recherche ont été lancés dans le monde entier pour améliorer la dynamique des cobots par rapport à la sécurité, ANR-SISCob (Safety Intelligent Sensor for cobots) étant l'un de ces projets. Les travaux menés au cours de cette thèse ont pour but de concevoir des dispositifs de sécurité qui sécuriseront les robots en y introduisant l’aspect de compliance. En effet, nous avons développé deux dispositifs dans lesquels l'aspect sécurité est atteint avec deux approches différentes :- Prismatic Compliant Joint (PCJ) : qui vise à la mise en œuvre dans les articulations linéaires, car peu de travaux ont traité de tels systèmes d'actionnement. Ici, la sécurité est atteinte biomimétiquement tout en faisant face à d'autres critères de sécurité liés aux propriétés mécaniques du corps humain.- Variable Stiffness Safety Oriented Mechanism (V2SOM) : Contrairement au premier dispositif d'inspiration biomimétique qui sert aux systèmes d'actionnement linéaires, le profil de sécurité du V2SOM est axé sur la sécurité selon deux critères de sécurité: force d’impact et HIC. L'aspect ‘orienté sécurité’ est dû à ce que nous appelons la capacité de découplage d'inertie de son profil de rigidité. V2SOM est actuellement dans ses dernières étapes de brevetage.Ces deux appareils seront intégrés dans un robot sériel réalisé dans notre laboratoire. / In the recent years, we witnessed a paradigm shift from making stiff robots toward compliant ones. This is due to several reasons such as enhancing the efficiency of robots in making explosive or cyclic motion. In fact, one of the earliest motivations from which this change stems are safety. Speaking of safety of both the human subject and the robot alike, while engaging in a collaborative task. Thus, the designation of cobots. Cobots may assist well-experienced human operator in several domains where precision is a must, such as industrial applications or medical tasks. Until now cobots still display safety concerns, even with regulatory recommendations such as ISO/TS 15066 and ISO 10218-1 et 2 that limits their economic benefits. In this view, several research projects were launched worldwide to enhance the cobot’s dynamics vs safety, ANR-SISCob (Safety Intelligent Sensor for cobots) is one of these projects. The works conducted during this thesis aims at making safety devices that will make robots safe by introducing compliance aspect in them. Indeed, we developed two devices in which safety aspect is achieved with two different approaches: - Prismatic Compliant Joint (PCJ): is aimed at prismatic joint’s implementation, as few works have dealt with such actuation systems. Herein, safety is biomimetically attained while coping with other safety criteria related to the mechanical properties of human body. - Variable Stiffness Safety Oriented Mechanism (V2SOM): Unlike the first device that’s biomimetically inspired and serves at linear actuation systems, V2SOM’s safety profile is safety oriented according to two safety criteria Impact force and HIC, and is designed for rotary actuation. The safety oriented aspect is due to what we call inertia decoupling capacity of its stiffness profile. V2SOM is currently in its final patenting process.Both devices will be integrated in serial robot built in our lab.

Les robots humanoïdes sont les plus appropriés pour travailler en coopération avec l'homme. En effet, puisque les humains sont naturellement habitués à collaborer entre eux, un robot avec des capacités sensorielles et de locomotion semblables aux leurs, sera le plus adapté. Cette thèse vise à rendre les robot humanoïdes capables d'aider l'homme, afin de concevoir des 'humanoïdes collaboratifs'. On considère ici la tâche de transport collaboratif d'objets. D'abord, on montre comment l'utilisation simultanée de vision et de données haptiques peut améliorer la collaboration. Une stratégie combinant asservissement visuel et commande en admittance est proposée, puis validée dans un scénario de transport collaboratif homme/humanoïde.Ensuite, on présente un algorithme de génération de marche, prenant intrinsèquement en compte la collaboration physique. Cet algorithme peut être spécifié suivant que le robot guide (leader) ou soit guidé (follower) lors de la tâche. Enfin, on montre comment le transport collaboratif d'objets peut être réalisé dans le cadre d'un schéma de commande optimale pour le corps complet. / Humanoid robots provide many advantages when working together with humans to perform various tasks. Since humans in general have alot of experience in physically collaborating with each other, a humanoid with a similar range of motion and sensing has the potential to do the same.This thesis is focused on enabling humanoids that can do such tasks together withhumans: collaborative humanoids. In particular, we use the example where a humanoid and a human collaboratively carry and transport objectstogether. However, there is much to be done in order to achieve this. Here, we first focus on utilizing vision and haptic information together forenabling better collaboration. More specifically the use of vision-based control together with admittance control is tested as a framework forenabling the humanoid to better collaborate by having its own notion of the task. Next, we detail how walking pattern generators can be designedtaking into account physical collaboration. For this, we create leader and follower type walking pattern generators. Finally,the task of collaboratively carrying an object together with a human is broken down and implemented within an optimization-based whole-bodycontrol framework.

Robots have growing potential to enter the daily lives of people at home, at work, and in cities, for a variety of service, care, and entertainment tasks. However, several challenges currently prevent widespread production and use of such human-centered robots. The goal of this thesis was first to help overcome one of these broad challenges: the lack of basic safety in human-robot physical interactions. Whole-body compliant control algorithms had been previously simulated that could allow safer movement of complex robots, such as humanoids, but no such robots had yet been documented to actually implement these algorithms. Therefore a wheeled humanoid robot "Dreamer" was developed to implement the algorithms and explore additional concepts in human-safe robotics. The lower mobile base part of Dreamer, dubbed "Trikey," is the focus of this work. Trikey was iteratively developed, undergoing cycles of concept generation, design, modeling, fabrication, integration, testing, and refinement. Test results showed that Trikey and Dreamer safely performed movements under whole-body compliant control, which is a novel achievement. Dreamer will be a platform for future research and education in new human-friendly traits and behaviors. Finally, this thesis attempts to address a second broad challenge to advancing the field: the lack of standard design methodology for human-centered robots. Based on the experience of building Trikey and Dreamer, a set of consistent design guidelines and metrics for the field are suggested. They account for the complex nature of such systems, which must address safety, performance, user-friendliness, and the capability for intelligent behavior. / text

Durant ce travail de thèse, nous nous sommes intéressés à la commande d'un robot manipulateur industriel, configuré pour une co-manipulation avec un opérateur humain, en vue de la manutention de charges lourdes. Dans un premier temps, nous avons présenté une vue d'ensemble des études qui ont été menées dans ce cadre. Ensuite, nous avons abordé la modélisation et l'identification des paramètres dynamiques du robot Denso VP-6242G. Nous avons utilisé le logiciel OpenSYMORO pour calculer son modèle dynamique. Après une présentation détaillée de la méthode d'identification des paramètres de robots manipulateurs, nous l'avons appliqué au cas de notre robot. Cela nous a permis d'obtenir un vecteur des paramètres qui garantit une matrice d'inertie définie positive pour n'importe quelle configuration articulaire du robot, tout en assurant une bonne qualité de reconstruction des couples pour des vitesses articulaires constantes, ou variables au cours du temps. Par la suite, nous avons détaillé les nouvelles fonctionnalités proposées pour le générateur de trajectoire en temps réel, sur lequel repose notre schéma de commande. Nous avons présenté une méthode d'estimation de la force de l'opérateur à partir des mesures de la force d'interaction entre le robot et l'opérateur, tout en tenant compte de la pénalisation de la force de l'opérateur afin d'avoir une image de cette dernière permettant de générer une trajectoire qui respecte les limites de l'espace de travail. Des tests du générateur de trajectoire simulant différents cas de figure possibles nous ont permis de vérifier l'efficacité des nouvelles fonctionnalités proposées. Le générateur permet de produire une trajectoire dans l'espace de travail tridimensionnel selon la direction de l'effort appliqué par l'opérateur, ce qui contribue à l'exigence de transparence recherchée en co-manipulation robotique. Dans la dernière partie, nous avons présenté et validé en simulation une commande en impédance dont les trajectoires de référence sont issues du générateur développé. Les résultats obtenus ont donné lieu à une bonne qualité de poursuite des trajectoires désirées. D'autre part, le respect des limites virtuelles de l'espace de travail a également été pris en compte. Cependant, les trajectoires articulaires correspondantes peuvent franchir les limites définies pour préserver l'intégrité du robot. / In this thesis, we were interested in the control of industrial manipulators in co-manipulation mode with a human operator for the handling of heavy loads. First, we have presented an overview of existing studies in this framework. Then, we have addressed the modeling and the identification of dynamic parameters for the Denso VP-6242G robot. We have used the OpenSYMORO software to calculate its dynamical model. After a detailed presentation of the method for identifying the robot's parameters, we have applied it to the case of our robot. This allowed us to obtain a vector of the parameters which guarantees a positive definite inertia matrix for any configuration of the robot, as well as a good quality of reconstruction of the torques in the case of constant joint velocities or in the case of variable ones over time. To continue, we have detailed the new features that have been proposed for the online trajectory generator, for which the control scheme is based on. We have presented a method for estimating the operator's force from the measurements of the interaction force between the robot and the operator, while taking into account for the penalization of the operator's force in order to have an information of this last which allows to generate a trajectory that respects the limits of workspace. Some tests of the trajectory generator simulating different possible scenarios have allowed us to check the effectiveness of the new proposed features. The generator makes it possible to produce a trajectory in the three-dimensional workspace according to the direction of the force applied by the operator, which contributes to fulfill the requirement of transparency that is sought in a co-manipulation. In the last part, we have presented and validated, in simulation, an impedance control whose reference trajectories are delivered by the proposed generator. The obtained results have shown a good trajectory tracking. On the other hand, the satisfaction of the virtual bounds of the workspace has also been nicely taken into account. However, the corresponding articular trajectories can cross the bounds defined to preserve the integrity of the robot.