Using Machine Learning Techniques in Real-World Mobile Robots

• Authors: Michael Kaiser;Volker Klingspor;José del R. Millán;Marco Accame;Frank Wallner;Rüdiger Dillmann
• Venue: IEEE Expert: Intelligent Systems and Their Applications
• Year: 1995

Applying machine learning techniques can help mobile robots meet the need for increased safety and adaptivity that real-world operation demands. The techniques also facilitate robot-to-user communication.Mobile robots are gradually leaving the laboratories and entering the real world. They are employed in automated factories, used for plant supervision, and are increasingly successful in service tasks such as health care. Almost 100 hospitals in the US and Europe use TRC's HelpMate, for example. However, if mobile robots are to find wide use in the real world, they must be able to learn from their experience. They must also be more adaptive, communicative, and safe.To meet these challenges, the complexity of the robot's control software must increase dramatically. The loop between the robot's perceptions and its actions must narrow on several levels of abstraction. This hierarchy of increasingly abstract situation-action rules must be grounded: it must map directly to the robot's basic sensing and motion capabilities. Building these rules requires that the knowledge related to both the task and the robot be codified.In addition, the robot's task specification must be accessible to the average user. The robot must learn to transform its own perceptions and actions into the user's language, and to compile the user's task specification into a set of directly executable commands. The omnipresent problems of symbol grounding and signal-to-symbol transformation become even more challenging if the robot is adapting its behavior according to both user needs and environmental changes.This article shows how machine learning techniques can help us meet these challenges. Using these techniques, we built increasingly abstract representations of a robot's perceptions and actions. This produced a symbolic description of what the robot knows and can do. Because this task is fairly complex, we first identified those subproblems that a learning method can solve efficiently, and isolated those with good classical solutions. Also, for a robot to solve a complex problem, we had to find solutions for several learning tasks. We identified these learning tasks and the learning techniques appropriate for their solution. To evaluate our approach, we used the mobile robots Priamos and Teseo.