Evolutionary software for autonomous path planning
- LLNL
This research project demonstrated the effectiveness of using evolutionary software techniques in the development of path-planning algorithms and control programs for mobile vehicles in radioactive environments. The goal was to take maximum advantage of the programmer's intelligence by tasking the programmer with encoding the measures of success for a path-planning algorithm, rather than developing the path-planning algorithms themselves. Evolutionary software development techniques could then be used to develop algorithms most suitable to the particular environments of interest. The measures of path-planning success were encoded in the form of a fitness function for an evolutionary software development engine. The task for the evolutionary software development engine was to evaluate the performance of individual algorithms, select the best performers for the population based on the fitness function, and breed them to evolve the next generation of algorithms. The process continued for a set number of generations or until the algorithm converged to an optimal solution. The task environment was the navigation of a rover from an initial location to a goal, then to a processing point, in an environment containing physical and radioactive obstacles. Genetic algorithms were developed for a variety of environmental configurations. Algorithms were simple and non-robust strings of behaviors, but they could be evolved to be nearly optimal for a given environment. In addition, a genetic program was evolved in the form of a control algorithm that operates at every motion of the robot. Programs were more complex than algorithms and less optimal in a given environment. However, after training in a variety of different environments, they were more robust and could perform acceptably in environments they were not trained in. This paper describes the evolutionary software development engine and the performance of algorithms and programs evolved by it for the chosen task.
- Research Organization:
- Lawrence Livermore National Lab., CA (US)
- Sponsoring Organization:
- USDOE Office of Defense Programs (DP) (US)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 12570
- Report Number(s):
- UCRL-JC-133191; YN0100000; YN0100000
- Country of Publication:
- United States
- Language:
- English
Similar Records
Synthesis of logic circuits with evolutionary algorithms
Some experiments in machine learning using vector evaluated genetic algorithms