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.

Cette thèse traite du problème de la chute de robots humanoïdes. L’étude consiste à découpler la stratégie de chute en une phase de pré-impact et une phase de post-impact. Dans la première, une solution géométrique permet au robot de choisir des points d’impact dans un environnement encombré. Pour ce faire, le robot réadapte sa posture tout en évident les singularités de chute et en préparant le seconde phase. La phase de post-impact utilise une commande par Programmation Quadratique (QP) qui permet d’adapter les gains Proportionnels-Dérivés (PD)des moteurs en ligne, ceci afin d’obtenir de la compliance dans les articulations. L’approche consiste à incorporer les gains de raideur et d’amortissement dans le vecteur d’optimisation du QP avec les variables habituelles que sont l’accélération articulaire et les forces de contact. Les contraintes ont été adaptées à ce nouveau QP. Enfin,comme la solution est locale, une commande de modèle prédictif sur un modèle simplifié du robot. A chaque pas du développement, plusieurs expériences et simulations ont été effectuées. / This thesis deals with the problem of humanoid falling with a decoupled strategy consisting of a pre-impactand a post-impact stages. In the pre-impact stage, geometrical reasoning allows the robot to choose appropriateimpact points in the surrounding environment –that can be unstructured and may contain cluttered obstacles,and to adopt a posture to reach them while avoiding impact singularities and preparing for the post-impact. Thepost-impact stage uses a quadratic program controller that adapts on-line the joint proportional-derivative (PD)gains to make the robot compliant, i.e. to absorb post-impact dynamics, which lowers possible damage risks.We propose a new approach incorporating the stiffness and damping gains directly as decision variables in theQP along with the usually-considered variables that are the joint accelerations and contact forces. By doing so,various constraints can be added to the QP. Finally, since the gain adaptation is local, we added a preview ona time-horizon for more optimal gain adaptation based on model reduction. At each step of the development,several experiments on the humanoid robot HRP-4 in a full-dynamics simulator are presented and discussed.