ICS Research Seminar on "Spatiotemporal prediction for bootstrap learning of robot sensory-motor capabilities"


"Spatiotemporal prediction for bootstrap learning of robot sensory-motor capabilities" by  Erhard Wieser, Chair for Cognitive Systems, TUM

Abstract: In developmental robotics, a long-term challenge is to make robots learn meaningful behavior and cognitive capabilities in an open-ended manner. A realization of such an open-ended learning system is very difficult and requires the implementation of a set of principles that are motivated from biology. I focus on a key principle inspired by the human cortex: spatiotemporal prediction. What is it, and how can it be useful for learning and generating sensory-motor capabilities on robots? I explain a particular type of spatiotemporal learner and suggest a way to optimize it. Moreover, I show how to extend it with other components, forming a cognitive architecture that learns to coordinate a robot. To this end, I explain the computational modelling, and show robot behaviors that result from particular components of the architecture, as well as sensory-motor capabilities that result from the entire architecture as a holistic system. I conclude by summarizing the benefits of my work.

For more info please visit:

15. June 2018, 10:00 h, Chair for Cognitive Systems, Karlstraße 45, 2. Floor, media room (2001)