Cognitive Robotics Workshop
Auke Jan Ijspeert
EPFL
Swiss Federal Institute of Technology, Lausanne
Luxembourg, December 20, 2005
Research interests
• To study/model the neural mechanisms
underlying adaptive locomotion and
movement control in animals
• To develop new control methods (e.g.
systems of coupled oscillators) for articulated
robots
• ANNs and dynamical systems approach
• Learning and optimization algorithms for
robotics
1) The potential for cognitive systems research of
platforms with right sensing and acting capabilities
• Embodiment: interacting with a (complex) environment,
acting on it, influencing its own perception (active
perception)
• Cognitive systems that can act and move in the real world,
reunifying locomotion/movement and cognition research
• Robots as tools for understanding animal functioning and
cognition
• Numerous potential applications for robots that can go out
in the real world: service, assistance/rehabilitation, rescue,
exploration/inspection, construction, Entertainment
2) Long term goals and leading edge research
Robot capable of moving as well as, or better than, animals
Adaptive locomotion
Coordination of
multiple degrees of freedom
Visuomotor coordination
Switching between motor tasks
Modulation
Learning new skills
2) Long term goals and leading edge research
Robots capable of moving as well as, or better than, animals
Robots that have internal models: that can predict their body
dynamics and the physics of the world, that can reason
about the world
Robots capable of learning new skills and adapting old skills
Robots capable of switching between different learning
frameworks (trial-and-error, by imitation, …)
Robots as tools to better understand animals
3) R&D goals and scope for interdisciplinary
co-operation
• Locomotion in complex and unstructured terrain (new
robots + new controllers). Going away from pre-recorded
trajectories to locomotion controllers that show limit cycle
behavior. Tight sensory-motor coordination
• Exploring the links between rhythmic movements (e.g.
locomotion) and discrete movements (e.g. reaching,
manipulation)
• Exploring the links between locomotion/movement
control and higher cognition
3) R&D goals and scope for interdisciplinary
co-operation (C’td)
• Possibly: decoding the evolutionary changes that led to
cognition
• Interdisciplinary cooperation: ethology, neuroscience,
biomechanics, sport science, mechanical engineering,
electrical engineering, control, material science, computer
science
4) Impact at different levels and in different areas
• Research:
• catching up with Japan and the US,
• give more weight to the importance of moving and
acting on the world,
• provide good basis for robots that can do interesting
things
• Neuroscience: to understand how locomotion and
movement have shaped who we are and how we think
• Society: robot assistants, rescue robots, service robots,
entertainment robots,…
• Industry: cf Toyota’s vision (2 cars + one robot per house)