Research at the Artificial Intelligence Lab

 All projects contribute, one way or other, to our central theme of understanding intelligence. The central concepts which form the basis of our approach are listed in figure 1, embodiment (the physical realization of agents), morphology, system-environment coupling, dynamics, and material properties. Our main research fields are indicated in the boxes with the rounded corners, namely biorobotics, learning and development, evolution and morphogenesis, and collective intelligence. Then, the scientific disciplines participating in the endeavor are listed as well as a number of important applications. Finally, there is a selection of robots developed at the lab and a number of applications.
 
 

Figure 1: Clickable Schematic of Research Program

 Biorobotics: inspiration from insects and humans

In order to explore the various aspects of intelligence a number of research strands are pursued (figure 1). In the biorobotics strand the goal is to model and learn from biological systems, such as the navigation behavior of insects, or the walking dynamics of humans. In particular we have been cooperating with the group of professor Rüdiger Wehner at the Zoology department of the University of Zurich. We have developed several robots (Sahabot 1, Sahabot 2, Analog Landmark Navigation Robot, Samurai) that mimic aspects of the navigation behavior of the desert ant Cataglyphis and have tested their performance in a salt pan in Southern Tunisia in Northern Africa. We have built a flying robot to explore the issues involved in 3-D insect navigation. This research is performed by David Andel, Dimitri Lambrinos , Ralf Möller, Verena Hafner, Fumiya Iida. In the same spirit we are working on modeling aspects of human behavior. In particular we are developing a biped walking robot for studying the autonomous learning potential of partially spinal cord injured patients. This work is performed by Chandana Paul and Hansruedi Frueh in cooperation with the Paraplegic Center of the University of Zurich (Balgrist), the group of Professor Volker Dietz.

 Learning and development: system-environment coupling and neural modeling

A core ingredient of intelligence is learning. Inspired by the development of human infants we are using robots to mimic the ontogenesis of categorization behavior and in the long term of concepts. This is achieved by endowing the robots with developmental mechanisms and have them interact with the (physical and social) environment for extended periods of time. This new research field is called Developmental Robotics. It turns out that in development embodiment (morphology of sensor and motor system) and processes of sensory-motor coordination play an essential role: they provide the "grounding" for the concepts. In the past we have been using the Samurai robots. We are currently switching to a Mitsubishi RV-2AJ, a small state of the art 5+1 DOF industrial robotarm. This work is performed by Massimiliano Lungarella and Gabriel Gomez. We are also cooperating with developmental psychologists and neuroscientists on this issue. We are currently in the process of setting up a cooperation with Sony Computer Science Laboratory in Paris in order to include communication into the developmental process. It is expected communication provides additional constraints on development that may speed up learning significantly.

Collective intelligence: emergence and self-organization

Intelligence never occurs in isolation but always in the context of societies. In our simulation models and robot implementations we are systematically exploring the emergence of structures in groups of interacting agents. We explore the resemblance of these artificial societies with those of real animals such as primates. This work and extensions of it is performed by Charlotte Hemelrijk, Rene te Boekhorst, and Hanspeter Kunz, part of it jointly with the Marine Research Lab in Woods Hole (Cape Cod, Mass.). In this context we are also exploring flocking using the Samurai robot platform. We are evolving dynamically rearranging neural networks to evolve a control architecture in cooperation with Akio Ishiguro of Nagoya University. This work is performed by Hanspeter Kunz (link to project).

 The origins of intelligence: evolution and morphogenesis

All forms of natural intelligence are the result of a process of evolution: they have evolved as physically embodied systems. Because of the fundamental importance of morphology for intelligence, we have been developing evolutionary methods that enable us to investigate aspects of morphology. In particular we are studying processes of morphogenesis, i.e. how the genotype of an individual is translated into the phenotype. The pertinent models closely mimic biological processes of cell-to-cell communication. We have applied these ideas to the generation of butterfly wing patterns, the evolution of the morphology of an insect eye, and to the generation of 3-D shapes in general. This work was started by Peter Eggenberger, and is continued by Josh Bongard, Daniel Bisig, Dale Thomas, Raja Dravid, and Lukas Lichtensteiger. The application of this work to changing morphologies and to the design of intelligent machines has lead to a project entitled "Morpho-functional machines" which is conducted jointly with the Science University of Tokyo (Fumio Hara). We are applying evolutionary methods to investigate morphology and control of a multi-segmented robot arm, work performed by Raja Dravid.
  An important focus is on physically realistic simulation so we can test the phenotypes from our evolutionary simulations. Another is on rapid robot building kits so that we can quickly test the evolved phenotypes in the real world.

 
Applications: bringing robot technology to the real world

As one core application we are cooperating with the Paraplegic Center of the University of Zurich (Balgrist) to investigate autonomous learning capabilities of spinal cord injured patients by building a biped walking robot (see above). Further applications are developed jointly with Starseed Enterprise, another spin-off company of the Artificial Intelligence Laboratory.

 Situated cognition: information seeking in the Internet

In addition to the projects mentioned above we are exploring the application of ideas of situated cognition-viewing cognition in context to information seeking tasks on the Internet. This work is performed by Christopher Lueg in the context of the European project SELECT

Go to AILab main page