Biped robot locomotion in scenes with unknown obstacles

The focus of this work is on development of techniques for sensor-based motion planning for a biped robot operating in a scene with unknown obstacles. While the planning and control for biped robots is an important and popular research area, little has been done in it on sensor-based planning and collision avoidance. One goal of this work is to fill this gap.; The specifics of motion planning for a biped robot is in that biped motion is inherently dynamically unstable. While normal stepping can be designed to assure dynamic stability, this pattern is not fit to negotiate (for example, step over) an obstacle. A change required in the step pattern becomes a disturbance that threatens stability. Hence the problem of special walk control for obstacle negotiation. This control may require deviation from a line motion. To our knowledge, this is the first work where biped locomotion becomes truly two-dimensional; the standard pattern in prior work is simple straight line motion.; Based on the methodology developed here and using on-line sensor information about its surroundings, a biped robot becomes capable of negotiating obstacles on its way. In each act of obstacle negotiation, depending on the obstacle's shape and location, the decision-making algorithm chooses from a list of carefully precomputed walking patterns a pattern that fits the immediate task best. The robot then negotiates the obstacle and resumes stable motion. Together, the set of precomputed patterns covers a reasonably wide range of possible situations.; As in most of prior work, our means for obtaining dynamic stability are based on the concept of Zero Moment Point (ZMP); ZMP is defined as a point of the walking surface in which the total forces and moment acting on the robot are zero. Another tool used is the strategy (from our own prior work) for generating biped stable locomotion trajectories under a side force disturbance.; Simulation experiments demonstrate stability of motion when negotiating various obstacles in local and in global settings. The approach is fast enough for real-time implementation.