Description and Design of Motion Systems for Autonomous Robots based on a Real-Time Parallel Computational Model

Abstract

The brain of an autonomous robot generates signals to the motors in each situation of the dynamical real world to achieve various tasks. It will be a complex system that maps the sensation and the intention of the robot into the commands to the actuators, when the robot makes many actions and has many degrees of freedom. In this paper, we propose to describe a hierarchical and modular motion system as a BeNet, an asynchronous network of real-time computational modules that represents a dynamical system completely, which interacts with the physical world. The designer decomposes the system into parallel modules which have distinct roles, and builds an understandable and desirable system, based on our abstract computational model. We show how to design a hierarchical BeNet for desirable actions of an autonomous legged robot, utilizing our programming environment. The system consists of pattern generators and action generators whose roles are stipulated so that the system is easy to debug or extend. In our environment, we can describe a modular system and verify it immediately with a real robot. Each module is described briefly in C and desirable actions of a robot with multiple DOF's were generated by the system.