Control and motion planning of dynamically walking bipeds for cooperative transportation
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
There is an increasing demand for assistive bipedal robots that are capable of physical interaction with other agents possibly humans to accomplish collaborative tasks, such as coordinated object transportation. In such collaborative scenarios, we can rely on the environment mapping and path planning skills of the leading collaborator to choose an obstacle-free trajectory for the team. This intended trajectory may not be directly accessible by the robot; however, the interaction forces developed between the robot and the collaborator offer cues on how the robot should adapt its behavior to accomplish the task. As its first objective this thesis proposes a method that empowers a biped to actively modify its speed and heading angle in response to the resulting interaction forces, allowing a collaborator to effectively walk the biped along a desired path. The proposed method is based on integration of impedance control to provide compliance at biped’s manipulator, with position control to synchronize the actuated degrees of freedom in a way that the generated walking gaits are adaptable to external activity. The feasibility of the method is illustrated on both planar and three dimensional bipedal robots that track the intended trajectory of leader with the sole knowledge of interaction force. ☐ Bipedal robots should also be capable of navigating an environment autonomously, that is without the help of a leading collaborator. Planning the motion of biped through a workspace populated by obstacles can be decomposed into two hierarchical components. At the high level, a planner is responsible for the generation of an obstacle free path that respects the geometry of the workspace. At the low level, a controller should take into account the stability of the platform as it executes the descending plan. Certain stability issues and unfaithful execution of the plan may arise if the high-level planning and low-level stability goals are considered in isolation. The second objective of this thesis is to bridge this gap by proposing a framework that unifies low-level stability and high-level planning objectives for systems that move in the environment via cyclic interactions, such as dynamically walking bipeds, in order to stably navigate them through a cluttered environment. The framework is based on extracting motion primitives in the form of limit cycle locomotion behaviors. The planner outputs a sequence of motion primitives that has to be followed by the robot in order to reach a goal location while avoiding obstacles. In this setting, a discrete-time switched system with multiple equilibria – each corresponding to a motion primitive – emerges as a natural formulation of the problem which projects the stability of the motion sequence to that of the switched system. We then show that the solution of the switched system can be confined in a safe region – characterized as the union of sub-level sets of Lyapunov functions – by imposing a bound on the dwell time of the switching signal. The approach is implemented on an underactuated 3D biped, and locally exponentially stable gait primitives are extracted using Hybrid Zero Dynamics (HZD) controllers. The dimensional reduction afforded by HZD allows the estimation of the basin of attraction of the gait primitives using sums-of-squares techniques, which facilitates the computation of the bound on the dwell time. ☐ Overall, this thesis contributes to the cooperation and autonomous navigation of dynamically (limit-cycle) walking bipedal robots in two ways. First, it takes a step toward the development of controllers for cooperative object transportation tasks, in which a bipedal robot assists a human to carry an object along a path that is enforced by the human. Secondly, it bridges the gap between high-level motion planning algorithms and low-level limit-cycle locomotion controllers so that the descending commands of the planner can be faithfully executed by the biped.
Description
Keywords
Applied sciences, Bipedal robots, Control systems, Cooperation, Humanoids, Legged locomotion, Motion planning