Investigation of a General Artificial Intelligence Framework for Robotic Control,
Inclusive Dates: 03/05/12 – 07/05/12
Background — As robotic systems become more complex, intelligent control must be applied to further advance the field of robotics. For more than two decades, SwRI has fabricated and maintained coating removal systems for large aircraft for the U.S. Air Force. These large robotic systems are currently controlled with a human operator monitoring video feedback of the coating removal process and adjusting the robot speed accordingly. A method of automating the speed control based on coating removal sensory feedback will result in greater throughput and higher reliability of these existing systems. However, traditional machine learning techniques are unlikely to be effective because of the complexity of the video data due to variance in lighting conditions, coating color and surface geometry. Therefore, this research focused on applying a general artificial intelligence (AI) framework to these robotic coating removal systems. The goal of this effort was to produce appropriate speed control decisions using AI given the input video data and to evaluate the chosen AI framework.
Approach — The AI framework was used to construct an intelligent "agent," which is the structure of how the data is passed around within the framework. The video data was first manipulated into a feature vector for the agent input. The agent was allowed to make its best guess on speed decisions based on the input feature vector. The agent was then trained by applying positive feedback for good speed decisions and negative for bad speed decisions. This feedback and performance information was stored into a knowledge base for future decisions. Finally, the agent was exposed to unseen video data to test the appropriateness of its speed control decision.
Accomplishments — The image processing algorithms produced for this agent were remarkably robust and produced a realistic feature vector for a variety of video conditions. These tools will be useful for future efforts to automate coating removal processes and for a variety of surface processing applications. These results have been shared with customers in the coating removal market and are under consideration for improving existing robotic systems. The general AI framework used for this application is a recent development that required some refinement efforts prior to testing. As a result of this effort, SwRI researchers now have an extensive understanding of the capabilities this framework and can accurately apply it to other future applications. The unique information theories and technique applied through the framework could be powerful for a variety of applications as it matures.