Automatic Programming of obots using

The goal in automatic programming is to get a computer to perform a task by telling it what needs to be done, rather than by explicitly programming it. This paper considers the task of automatically generating a computer program to enable an autonomous mobile robot to perform the task of moving a box from the middle of an irregular shaped room to the wall. We compare the ability of the recently developed genetic programming paradigm to produce such a program to the reported ability of reinforcement learning techniques, such as Q learning, to produce such a program in the style of the subsumption architecture. The computational requirements of reinforcement learning necessitates considerable human knowledge and intervention, whereas genetic programming comes much closer to achieving the goal of getting the computer to perform the task without explicitly programming it. The solution produced by genetic programming emerges as a result of Darwinian natural selection and genetic crossover (sexual recombination) in a population of computer programs. The process is driven by a fitness measure which communicates the nature of the task to the computer and its learning paradigm. Introduction and Overview In the 195Os, Arthur Samuel identified the goal of getting a computer to perform a task without being explicitly programmed as one of the central goals in the fields of computer science and artificial intelligence. Such automatic programming of a computer involves merely telling the computer what is to be done, rather than explicitly telling it, step-by-step, how to perform the desired task. In an AAAI-91 paper entitled “Automatic Programming of Behavior-Based Robots using Reinforcement Learning” Mahadevan and Connell (1991) reported on using reinforcement learning techniques, such as Q learning (Watkins 1989), in producing a program to control an autonomous mobile robot in the style of the subsumption architecture (Brooks 1986, Connell 1990, Mataric 1990). In particular, the program produced by reinforcement learning techniques enabled an autonomous mobile robot to perform the task of moving a box from the middle of a room to the wall. In this paper, we will show that the James P. Rice Stanford University Knowledge Systems Laboratory 70 1 Welt h Road Palo Alto, CA 94304 USA [email protected] 415-723-8405 onerous computational requirements imposed by reinforcement learning techniques necessitated that a considerable amount of human knowledge be supplied in order to achieve the reported “automatic programming” of the box moving task. In this paper, we present an alternative method for automatically generating a computer program to perform the box moving task using the recently developed genetic programming paradigm. In genetic programming, populations of computer programs are genetically bred in order to solve the problem. The solution produced by genetic programming emerges as a result of Darwinian natural selection and genetic crossover (sexual recombination) in a population of computer programs. The process is driven by a fitness measure which communicates the nature of the task to the computer and its learning paradigm. We demonstrate that genetic programming comes much closer than reinforcement learning techniques to achieving the goal of getting the computer to perform the task without explicitly programming it. Background on Genetic Algorithms John Holland’s pioneering 1975 Adaptation in Natural and Artificial Systems described how the evolutionary process in nature can be applied to artificial systems using the genetic algorithm operating on fixed length character strings (Holland 1975). Holland demonstrated that a population of fixed length character strings (each representing a proposed solution to a problem) can be genetically bred using the Darwinian operation of fitness proportionate reproduction and the genetic operation of recombination. The recombination operation combines parts of two chromosome-like fixed length character strings, each selected on the basis of their fitness, to produce new offspring strings. Holland established, among other things, that the genetic algorithm is a mathematically near optimal approach to adaptation in that it maximizes expected overall average payoff when the adaptive process is viewed as a multi-armed slot machine problem requiring an optimal allocation of future trials given currently available information. Recent work in genetic algorithms can be surveyed in Goldberg 1989 and Davis (1987, 1991). 194 Learning: Robotic From: AAAI-92 Proceedings. Copyright ©1992, AAAI (www.aaai.org). All rights reserved.