| Summary: | 碩士 === 國立高雄應用科技大學 === 機械與精密工程研究所 === 95 === In order that a biped robot can walk smoothly on regular terrains and adapt easily to irregular
terrains or unexpected disturbances, lots of algorithms have been proposed in literatures. The most
feasible one is to plan the stable walking patterns for regular terrains in advance and modify them in
real-time when the robot is walking in those circumstances different to the planned one. For the same
purpose as above-mentioned, the dynamic model of a robot is simplified herein as a simple inverted
pendulum, and some feasible algorithms are proposed.
Firstly, a fast algorithm is adopted to generate the stable walking patterns for regular terrain, by
specifying the smooth ZMP trajectories and the parametric surfaces for the center of gravity (COG) of
the robot. Since it is more desirable to control a robot to walk stable with less energy and natural like a
human, a genetic algorithm is applied to adjust the design parameters of walking patterns such that
desired optimal patterns are determined.
Secondly, in order to control a biped robot to perform those motions other than walking, a
graphical-based interactive motion editor is developed. This motion editor is used to define the key
postures for the desired motion of a robot in time sequences, by specifying its joint angles. The
continuous postures for the desired stable motion are then generated by interpolating additional
postures in between the specified key postures. In the interpolation process, the stability of the
generated postures is ensured by using the genetic algorithm to adjust the additional postures to satisfy
the ZMP criteria.
Thirdly, an on-line modification for the postures to prevent the robot from tipping over is
proposed. The basic concept for keeping the stable robot motion is to modify the most effective joint
angle(s) to change the position of the COG of the robot. For a robot in motion, the difference between
the actual ZMP and the planned one can be determined by using sensors, and the difference should be
reduced for stable motion. Thus, the difference between the ZMP is first transformed into the difference
between the actual COG and the planned one, and then the so-called Jacobian method is applied to
determine the most effective joint angle(s).
Finally, experiments of walking and kicking motions are made for the proposed methods to
control a humanoid robot. The experiments show that the methods are feasible and reliable.
|
|---|
