Proceedings of a conference held in Bristol, 26th March 1999. U. Nehmzow and C. Melhuish (eds.)
One attraction of mobile robotics is the diversity of skills, techniques and approaches required to make a mobile robot behave intelligently in its target environment. There are hardware aspects, such as actuator and sensor design, there are engineering aspects such as sensor signal interpretation and control, and there are questions related to robot control paradigms, Artificial Intelligence and Cognitive Science applications in mobile robotics, such as learning, navigation and modelling of robot-environment interaction.
TIMR 99 reflects all these aspects, covered in a wide range of papers. These papers serve as pointers to problems that are currently being discussed within the mobile robotics community, and indicate open questions and directions of future research. Their main purpose, however, is to stimulate discussion.
Welcome to Bristol, and wishing you a successful and stimulating conference!
In this work a new technique for minimising jerk during longitudinal manoeuvre of a wheeled mobile robot traversing fixed road obstacles, which have positive gradient geometry, is presented. The dynamics of the robot with respect to the longitudinal, pitching and vertical motions of the platform are considered. The work has demonstrated that by controlling the drive torque input to the wheels, the platform can move over a positive-gradient hump with minimal jerk. This helps to minimise the possibility of wheel-ground contact loss, thus ensuring the stability of payload on wheeled mobile robots which are often used for transporting materials over uneven factory shop floors.
Learning systems in the field of mobile robotics are now quite commonplace and although they generally prove to be competent at the tasks set for them, their learning algorithms remain static rather than adapting through experience. Recently, ``lifelong'' learning systems have begun to appear that, over time, improve the way they learn, signifying the next generation of learning systems. This paper describes one such proposed system which is currently being developed. The chosen hardware and software components are discussed, including details of the schema-based learning system, and the expected behaviour of the agent is described.
This paper provides an overview of a three year project conducted
on the subject of self-localisation and autonomous navigation by a
The research was carried out in two stages:
Self-Localisation. Due to the fundamental unreliability of dead reckoning, landmark-based methods were investigated. Several related issues were addressed, including performance evaluation, replication and comparison of existing work, and the development of a novel localisation system based on the results obtained.
Autonomous Navigation.This work included the development of novel mechanisms for exploration and map building, as well as validating the previous work on self-localisation, in a complete, navigating mobile robot. Quantitative assessment of the different competences required for navigation was also carried out.
This paper presents a modular mapping model that allows an autonomous mobile robot with inaccurate sensors to construct a map of an unknown, unstructured and unmodified environment. The robot constructs the map in real-time autonomously with no a priori knowledge. To achieve this topological and geometrical methods are brought together to produce a mapping model that is robust, efficient and modular with a flexible method for sensor fusion. The produced map being adequate for path planning and localisation tasks. Moreover, it is believed this method achieves these tasks more elegantly than other existing methods. Experiments are currently being conducted on a mobile indoor robot. The experiments will later be extended to outdoor robots. The inspiration for this model was found in the biological literature.
This paper describes a telerobotic system designed for the real world task of bomb disposal. The system is based around an existing remotely operated vehicle built for this purpose some years ago. The original vehicle has been enhanced in order to contend with a number of operational problems identified with its use. A sensor configuration and distributed control system are described which address these problems. The core of the control system is founded upon the principles of the Subsumption Architecture, although the design is grounded in a wider concept of the development of generic user interfaces which may be used to control a number of different types of telerobot. This is a desirable feature for application domains in which a number of different types of robot are used for performing distinct functions, such as the nuclear industry and military operations. Localising the differences between telerobots to within the remote control system can increase flexibility and can simplify the development of different versions of a system. This requirement presents a number of problems when considering the physical structure (sensors, chassis and actuators) and the control systems (functions and behaviors) of different telerobots. To this end, a system architecture, communication protocol and robot description language are described which address these issues.
This paper presents a series of experiments with an autonomous mobile robot to acquire a model of the robot's interaction with a specific environment. In these experiments, an artificial neural network model of the environment is acquired, using data obtained from the robot interacting with the real world. To investigate the feasibility of this approach we have modelled fundamental sensory perception, robot environment interaction in simple, small environments, and the application of this novel approach to a machine learning application in robotics. The behaviour of a learning mobile robot is predicted using our network-based approach and shown to produce more faithful results than classical robot simulation methods.
Habituation, the mechanism by which the brain learns to ignore repeated stimuli, is the most basic form of plasticity within the brain. It has been investigated thoroughly in recent years by biologists in a number of different animals, from the sea snail Aplysia through to cats and toads. Of itself, habituation is an interesting phenomenon and therefore worthy of investigation, but there are several worthwhile uses to which it can be put. For example, a novelty detector of the type proposed by Kohonen is effectively an habituation mechanism since input patterns which are presented to the network repeatedly are ignored.
Intelligent mobile robots need the ability to integrate robust navigation facilities with higher level reasoning. This paper is an attempt at combining results and techniques from the areas of robot navigation and of intelligent agency. We propose to integrate an existing navigation system based on fuzzy logic with a deliberator based on the so-called BDI model. We discuss some of the subtleties involved in this integration, and illustrate it on a simulated example. Experiments on a real mobile robot are under way.
This paper reports on the preliminary findings of the application of collective minimalist robotics to the task of wall building. In particular it shows how linear wall structures, composed of circular objects, can be formed by a group of simple mobile robots without recourse to direct communication, sophisticated sensing or intensive computation. Studies on social insects, such as ants, reveal that construction of complex nest structures are built by animals with no internal 'blueprint' of the global structure. Such structures emerge as the consequence of the interactions of the insects, carrying out simple behaviours, and the environment. In particular, the exploitation of heterogeneities, referred to as templates, is investigated. Such a biological paradigm involving simple 'agents' limited in their capacity for computation, communication and sensing has an obvious appeal to those researching into minimalist collective robotics. The underpinning biological mechanisms may be of particular importance when building very small robots where computation, sensing and ability to communicate may be seriously constrained.
This paper advocates the application of multi-agent techniques in the realisation of social robotic behaviour. We present an architecture which commissions agent-based deliberation without sacrificing the reactive qualities necessary in a real world domain, and which is situated within a social landscape through the use of an Agent Communication Language.
A novel method for determining the vanishing point in the television image of a corridor using a polar histogram is presented. A method is also proposed for using the angles of lines that converge on the vanishing point to generate steering signals to enable control of the lateral displacement of a robot travelling down a corridor. Results are presented which show that the method is robust with real images in the presence of noise, visually cluttered images and where some of the expected features are missing.
There are many mechanisms with which agents in multi-agent systems can communicate. These mechanisms can be described in various ways, depending on the aims of those running the system concerned. This paper proposes a classification system for such communication methods so that they can be described in a standard way and be more easily compared to other such systems. Part of the classification consists of a coded definition, facilitating automisation of some development.
Heron of Alexandria in around 60 A.D. described the construction of a self-moving cart. This wheeled base was moved by ropes wound round axles and attached to a counter-weight. By judicious winding of the rope the cart could be 'programmed' to move in any pattern. Other mechanisms described by Heron could have been used to make the cart reactive. This early example of an automaton influenced the development of a mechanistic view of behaviour, and is an interesting precursor to current mobile robots.
Unsupervised learning and supervised remote teleoperator control for robots may seem an unlikely combination. This paper argues that the combination holds advantages for both parties. The operator would like to "instruct" the robot without any special effort, and then be able to hand over some or all of the tasks to be performed without loss of overall supervisory control. In return, the learning algorithm receives a continuous stream of exemplar data relevant to the tasks it might later be asked to perform. We consider an unsupervised learning method, the Dynamic Expectancy Model, and a teleoperator "architecture" offering just such a serendipitous combination.