Seminar by M. Hoffmann: From locomotion to cognition
The next L&M seminar in Japan will take place next week, Friday Oct. 7, at 4:30pm in Room 107, Building 16, at the Komaba campus of the University of Tokyo. We are pleased to have Matej Hoffmann from Rolf Pfeifer’s AI Lab present us about his current research:
From locomotion to cognition
Artificial Intelligence Laboratory
University of Zurich, Switzerland
The cognitivistic paradigm, which states that cognition is a result of computation with symbols that represent the world, has been challenged by many. Artificial systems based on this paradigm have achieved many successes, but encountered also fundamental problems. A prominent example is the symbol-grounding problem: if symbols are implanted into the brain of an artificial agent, how is it supposed to relate them to the real world such that they become meaningful? Alternatives were provided by embodied cognitive science and behavior-based robotics where the importance of embodiment – interaction with the environment through a body – is emphasized. While such agents that do not rely on representations demonstrate some remarkable capabilities, they are still restricted to tasks of limited complexity. In our research we attempt to bridge this gap. In a series of case studies centered around a quadruped robot, we will demonstrate how the robot can gradually become more “cognitive” and how this brings behavioral advantage to the robot. The robot will autonomously develop minimal representations of its body and its interaction with the environment (a body schema or self-model), including a brief look-ahead in the form of a forward model. These primitive representations are formed in a multi-modal sensory-motor space and are thus firmly grounded in the body-environment dynamics.