From Neuroscience to Robotics: An Epigenetic Robotics Approach

Christian Balkenius & Birger Johansson

Lund University, Sweden

How can a robotic system be designed that develop in a way that resembles that of the young infant? This series of lectures will try to integrate data from cognitive and computational neuroscience with developmental psychology to suggest ways of designing robots that can autonomously develop cognitive skills.   In addition to the lectures, there will be five group sessions that will offer hands-on experimenting with the Ikaros system and real robots. These will follow the lectures and consist of practical experiments with computational and robotic models of early visual processing, learning and visual attention, anticipatory mechanism and goal directed behavior.

Course Outline

Lecture 1: Low-Level Sensory Processing

How does the brain code sensory signals, what is coded, and how does such codes develop from experience? Common principles in different sensory modalities. How can this be implemented in robots?

Required Readings:

Balkenius, C. (2007). Principles of Brain Design (Chapter 2: Neural Coding and Transformation), in prep.

Optional Readings:

Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-Based Learning Applied to Document Recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, Nov. 1998.

Bednar, J. A: and Miikkulainen, R. (2003). Learning Innate Face Preferences. Neural Computation 15(7): 1525–1557.

Lecture 2: High-Level Sensory Processing

How are complex stimuli recognized by the brain? How can an attention system be tuned by interaction with an environment? How does sensory processing interact with motivation and emotion?

Required Readings:

Balkenius, C. (2000). Attention, habituation and conditioning: toward a computational model. Cognitive Science Quarterly, 1, 2, 171-214.

Optional Readings:

Balkenius, C. (1995). Natural Intelligence in Artificial Creatures (Chapter 6: Motivation and Emotion).

Lecture 3: Reactive Behavior and Feedback

The role of reactive control in behavior. What is the role of hierarchies, parallel motor systems, and feedback in the control of actions? Interaction between attention and behavior.

Required Readings:

Balkenius, C. (1995). Natural Intelligence in Artificial Creatures (Chapter 3: Design Principles & Chapter 4: Reactive Behavior).

Lecture 4: Mechanisms of Learning and Development

How are new behaviors acquired? What is the role of rewards in learning? How does early learning influence later learning? How can increasingly more complex behaviors develop?

Required Readings:

Balkenius, C., and Moren, J. (1999). Dynamics of a classical conditioning model. Autonomous Robots, 7, 41-56

Optional Readings:

Singh, s. Barto, A G., Chentanez, N. (2004). Intrinsically Motivated Reinforcement Learning

Winberg, S. and Balkenius, C. (2007). Generalization and Specialization in Reinforcement Learning, manuscript.

Lecture 5: Interaction in Context

How do new behaviors develop through the ongoing dynamic interaction with the world? Moving from reactive to anticipatory behavior. How does the internal and external context structure behavior?

Required Readings:

Balkenius, C. and Johansson, B. (2007). Anticipatory Models in Gaze Control: A Developmental Model. Cognitive Processing, in press.

Optional Readings:

Balkenius, C., Moren, J. (2000). A Computational Model of Context Processing- Jean-Arcady Meyer, Alain Berthoz, Dario Floreano, Herbert L. Roitblat, Stewart W. Wilson, (Eds) From Animals to Animats 6. MIT Press

Balkenius, C., and Winberg, S. (2004). Cognitive modeling with context sensitive reinforcement learning. In Proceedings of AILS '04. Lund: Dept. of Computer Science.

Extra Material:

Balkenius, C., Moren, J. and Johansson, B. (2007). System-Level Cognitive Modeling with Ikaros, Lund University Cognitive Studies, 133.

Balkenius, C. & Prince, C. G. (2006). Tutorial: Featural Processing of Auditory and Visual Inputs for Epigenetic Robots.

Christian Balkenius

Lund University, Sweden

Christian Balkenius is Associate Professor of Cognitive Science at Lund University. He works with systems-level computational models of cognitive subsystems in the  brain. His work focuses on long-term learning and development and the relation between sensory coding, generalization, context and attention. In 1997-2000 he developed the first computational model to integrate context processing, inhibition and conditioning that could explain the results of a wide range of learning experiments. In 2001, he initiated the Ikaros project that aims at developing an open infrastructure for system-level brain modeling. He has also developed models of learning processes in the control of visual attention and neural interfaces for the control of an artifical hand. He has published some 100 research papers on neural network based modeling of cognitive processes, robotics, vision and learning theory. He is currently working on models of anticipation in the control of visual attention. He has been the organizer of seven workshops on Epigenetic Robotics.