Locomotion and balance control of humanoid robots with dynamic and kinematic constraints

Building a robot capable of servicing and assisting people is one of the ultimate goals in humanoid robotics. To realize this goal, a humanoid robot first needs to be able to perform some fundamental locomotion tasks, such as balancing and walking. However, simply performing such basic tasks in static, open environments is insufficient for a robot to be useful. A humanoid robot should also possess the ability to make use of the object in the environment to generate dynamic motions and improve its mobility. Also, since humanoid robots are expected to work and live closely with humans, having human-like motions is important for them to be human-friendly.;This dissertation addresses my work on endowing humanoid robots with the ability to handle dynamic and kinematic constraints while performing the basic tasks in order to achieve more complex locomotion tasks. First, as a representative case of handling dynamic constraints, a biped humanoid robot is required to balance and walk on a cylinder that rolls freely on the ground. This task is difficult even for humans. I introduce a control method for a humanoid robot to execute this challenging task. In order for the robot to be able to actively control cylinder's motion, the dynamics of the cylinder has been taken into account together with the dynamics of the robot in deriving the control method. Its effectiveness has been verified by full-body dynamics simulation and hardware experiments on the Sarcos humanoid robot. Second, as an example of tasks with kinematic constraints, I present a method for real-time control of humanoid robots to track human motions while maintaining balance. It consists of a standard proportional-derivative tracking controller that computes the desired acceleration to track the given reference motion and an optimizer that computes the optimal joint torques and contact forces to realize the desired acceleration, considering the full-body dynamics of the robot and strict constraints on contact forces. By taking advantage of the property that the joint torques do not contribute to the six degrees of freedom of the floating base, I decouple the computation of joint torques and contact forces such that the optimization problem with strict contact force constraints can be solved in real time. In full-body simulation, a humanoid robot is able to imitate various human motions by using this method.;Through this work, I demonstrate that considering dynamic and kinematic constraints in the environment in the design of controllers enables humanoid robots to achieve more complex locomotion tasks, such as manipulating a dynamic object or tracking given reference motions, while maintaining balance.