Synthesis of Comprehensive Connectionist Control Algorithms for Robot Contact Tasks

Abstract

The previously defined connectionist learning control laws for robotic contact tasks did not particularly deal with the problem of environment models and environment uncertainties. The main intention was to compensate for the system uncertainties by connectionist structures in a global fashion, without separately considering the problems of environment modelling, identification, and uncertainties. However, it is well known that most manipulation robots, especially those in industrial practice, are required to operate in uncertain and variable environments. Thus, the characteristics of the environment can be assumed to be unknown and to change significantly, depending on the given task. It is clear that without adequate knowledge about the environment dynamics it is even not possible to determine consistent values of nominal trajectory and force, as well as nominal control, not to mention achieving asymptotic stability. The previous research works in most cases, however, did not consider these uncertainties and nonlinearities of the environment and, thus, the above approaches are limited to specified working conditions, which satisfy only certain assumptions. Namely, when considering specific contact tasks, simplifications in the modelling of robot and environment dynamics are introduced in almost all control approaches, in order to obtain simpler control algorithms. Hence, the uncertain nonlinearities and other characteristics of environment models still remain a critical issue in robotic contact task research.