Flexible and Robust Walking in Uneven Terrain

点击翻译按钮获取中文摘要

There have been remarkable advances in the technology of biped robots, enabling the most advanced machines to walk and run at relatively high speeds. Still, current bipeds are not capable of handling the large disturbances that would occur during the practical use of such machines and still have severe limitations in the generation of walking patterns in complex environments. Currently, these limitations appear to be more problematic than the fact that robots can not walk or run as fast in ideal laboratory conditions as humans can, since a central argument for legged locomotion is the supposed superiority to wheeled or tracked vehicles in rough terrain. Therefore, the goal of this project is to research methods for improving the robustness and flexibility of biped robots. We view this research as a first step towards using biped robots outside tightly controlled laboratory conditions and proving the inherent superiority of legged locomotion in rough terrain. The project will include theoretical studies, numerical simulations and walking experiments using the biped robot Lola developed during the DFG project „Natur und Technik intelligenten Laufens“.The term „flexiblility“ in the title of the project signifies the capacity of the control system to actually use a large portion of the robot‘s physical capabilities. This is important, e.g., when stepping over or onto complex, previously unknown obstacles or when walking with very large steps. By term ‘robustness‘ we mean the ability of the robot to recover from very large disturbances, due to external forces or errors in the environment model. This can only be achieved by large modifications of the originally planned motion of the system. This proposal is based on two hypotheses: 1. Real-time planning of stable locomotion with singularity and collision avoidance is simplified and sped up by using a time-varying definition of task-space and a predictive inverse kinematics algorithm.2. Robustness against disturbances is increased by reactive re-planning of walking patterns and adaptive robot and environment models. The methods developed during this project are easily transferable to similar biped robots and, to a large extent, more conventional robot manipulators. Specifically, we are planning to use the methods developed in this project for motion generation of agricultural robots in cluttered, unknown environments at a later time.