The Oxford Robot World Model 1. Model Based Robotics

After reviewing the advantages of supplying a robot with a geometric model of its surroundings, we discuss the design of a modular object-oriented database to support such a model, which is being implemented as part of the Oxford Autonomous Guided Vehicle project. The Oxford Autonomous Guided Vehicle 1, 2] is a mobile robot truck or forklift designed to operate with a degree of autonomy in a structured and fairly well known factory-like environment. It has a number of advanced sensor systems for navigation, obstacle avoidance and pallet acquisition, and a corresponding set of task and path planning strategies, but at the heart of the software is a model-based representation of the vehicle and its environment, supported by a purpose-built geometric database. There are several good reasons for providing a robot with such a world model: (i) It provides an eeective, easy to use, high level interface to the robot: task level behaviour such as`pick up the pallet at x and carry it to y' is natural to a robot whichùnderstands' its surroundings. (ii) It encourages a correspondently clean internal architecture in which independent vehicle subsystems such as sensing and data fusion, vehicle controlling, and path and task planning modules communicate with one another mostly via the world model. (Albeit there are usually a few time or safety critical mechanisms such as basic vehicle reeexes which must bypass the model). This modular, inherently parallelisable, model-based, data-driven architecture simpliies the introduction of new subsystems by providing a uniform inter-module interface; in particular it allows vehicle capabilities to be extended by the addition of more or less genericèxpertise modules' which need only interface with the relatively simple and vehicle-independent model database. (iii) It decouples essentially robot-independent environmental information from the robot-speciic software, allowing it to be handled by eecient specialist code such as the spatial database and shared between a number of robot and control processes. The result is an integrated model of the whole environment which is immediately exten-sible to large multi-robot systems and provides a solid foundation on which to build interacting-robot and factory-wide controllers.