No abstract

A hierarchical architecture is described which supports space station telerobots in a variety of modes.T h e 8ystem is divided into three hierarchies: task decomposition, world model, and sensory processing.Coals at each level o f the task decomposition hierarchy a r e divided both spatially and temporally into simpler c o u a n d s for the next lower level. T h i s decomposition is r8peated until.at the lowest level, the drive signals t o the robot actuators are generated.to accomplish its goals, task decomposition modules must often u 8 e information stored in the world model.T h e purpose of the sensory system is t o update t h e world model as rapidly as possible to keep the model in regi8tratioa with the physical world.s h e architecture of the entire control syatem hierarchy and h o w it can be applied to space telerobot a p p l i c a t i o n s q m 4, ( , I~tRODUCTIOWOne of the major directions on which the robot research community has concentrated i t 8 efforts is concerned with planning and controlling motion. Given a specific task, a motiom pl.n mU8t be calculated which meets the task requirements. Then, the plan must be executed3 there must be sufficient control for the robot to adequately effect the desired motion. Trajectoriea are often planned as straight lines i n Cartesian space (1).Whitney (2.3) developed the resolved motion rate control method for Cartesian straight line motion8. Paul (4.5.6) umed homogeneous coordinate transformations to describe a trajectory as a function of time, and Taylor (7) used coordinated joint control over small segments to k e e p the trajectory within a specified deviation of the desired Straight line trajectory.While the teSe8rCh described above employs a -kinematic-approach to robot control. another direction of research takes the manipulator 'dynamics' into account in t h e description of robot motion.The dynamic equations of motion are described either by t h e Lagrangian formulation ( 8 ) or by the Newton-Buler equations (9).Algorithms and computer architectures have been suqge8ted which promise real-time dynamic robot control (10,ll). Operating with the constraints impoaed by real-time robot control, early methods used 8tructured light and binary images (13,14,15,16). Another aspect o f motion control is concerned with the variables being controlled. T h e te=eacch described to t h i s pointTheae approaches, though developed at different institutions, shared many concepts.One of the important subsequent research efforts went toward the development o f , model-based image procesaing. Bolles and Cain (17) used models of objects to guide t h e algorithms i n a hypothesis/verification scheme known a s the local feature focus method. T h e concept has recently been extended from two dimensional (i.e. nearly flat) objects to three dimensional objects (18).Although the approaches described here have led to a better underatanding of real-time vision procesaing, the systems lacked a sophiaticated interconnection wit'h the robot control system. 155https://ntrs.n...

No abstract

No abstract

No abstract