A hunt and gather robot

摘要

This thesis describes the development and implementation of a visually guided robotnthat learns to perform a task best described as qhunt and gatherq. The hunt and gathernoperation involves the negotiation of an environment in search of particular objectsn(hunting), followed by a retrieval of those objects (gathering). This behaviour forms thenbasis of a wide variety of dirty, dull and dangerous jobs that would be ideal to assign tonrobots; tasks such as fruit picking, weeding, domestic or industrial cleanup, or planetarynexploration.nnnnnnnnnnnnnnnn CORGI, the robot described in this thesis, learns how to perform a simple hunt andngather operation: finding tennis balls in my office. CORGI has only one sensor, a CCDncamera that provides visual information. Using neural network techniques, both perceptionnand control systems are developed, trained and implemented on the robot platform.nThe resultant system is capable of performing the hunt and gather task in real time:nusing only a single $20 microprocessor.nnnnnnnnnnnnnnnn The thesis reviews recent literature in robotics and neural networks in the context ofnBraitenberg's book Vehicles: An Experiment in Synthetic Psychology. The literaturenreview shows that many of Braitenberg's gestalt experiments are practicable to investigatenwith recent developments in robotic and neural network technology. A simulation,nSimCORGI, is used to show that the simplest of networks and training algorithms usednin neural network research can be used to train control systems for Braitenberg vehicles.nFurthermore, a perception system that is trained to detect drink cans is demonstrated in anpilot study to be suitable as a sensor system for a Braitenberg vehicle.n CORGI was developed based on the findings from these pilot studies. The robot hasna single Motorola MC68HC 1 6 microcontroller to perform all of the control, vision processing, communication and peripheral tasks required of the project. The robot's perceptionnsystem was developed by transmitting images from the robot to another computer,nwhere the images were tagged for supervised training based on thenback propagation algorithm. Investigations related to network architecture, and data representationnand presentation led to the development of perception systems suitable fornthe hunt and gather operation.nnnnnnnnnnnnnnnnnn The trained perception networks were then loaded back to robot for real time non-learningnperception. A behaviour based control system was developed based on the principlesninvestigated in simulation and run in real time on board the robot. The resultantnsystem is robust to changes in the environment, and readily generalises its behaviour tonnovel situations and unseen environments. The robot is also used in a real time behaviourntraining experiment, in which the robot learns the hunt and gather operation in fournminutes of on-board training.nnnnnnnnnnnnnnn Finally, the thesis reviews the approach to the problem and various aspects of thenimplementation. Further improvements and extensions to many aspects of the projectnare suggested.n