|
| EPSRC Reference: |
GR/R95722/01 |
| Title: |
Robot control using a model of central structures in the vertebrate brain |
| Principal Investigator: |
Professor T Prescott |
| Other Investigators: |
|
| Researcher Co-investigator: |
|
| Project Partner: |
|
| Department: |
Psychology |
| Organisation: |
University of Sheffield |
| Scheme: |
Standard Research |
| Starts: |
01 September 2002 |
Ends: |
31 August 2005 |
Value (£): |
161,244
|
| EPSRC Research Topic Classifications: |
| Brain Sciences |
Learning Engineering Systems |
|
| EPSRC Industrial Sector Classifications: |
| Electronics |
Healthcare |
| Information Technologies |
|
|
| Related Grants: |
|
| Panel History: |
|
|
Summary |
|
The design of control architectures that can generate integrated behaviour in multi-tasking, multi-actuator robots remains a central issue for researc autonomous robotics. There is also an increasing awareness in the robotics research community that an effective strategy for solving complex problem in robot design is to 'reverse-engineer' the biological control systems that underlie animal behaviour. The goal of this project is to develop a biomime 'integrative core' for robust and effective robot control by investigating candidate control architectures with embedded components modelled on key structures in the vertebrate brain. We will focus on two specific brain systems that are hypothesised to play an important role in action selection-the basal ganglia and the reticular formation. A series of novel robot control architectures will be developed based around biologically-accurate models of these structures and alternative hypotheses concerning their roles and interactions within the full functional architecture of the brain. Each candidate architecture will be evaluated on benchmark tasks designed to emulate aspects of animal foraging behaviour. Two important issues in contemporar robotics to be specifically addressed are the trade-off between heterarchical and hierarchical control, and the benefits of layered control architecture The development of embedded models of the basal ganglia and reticular formation should also have important implications for understanding the ro these structures in human brain function and dysfunction.
|
| Final Report Summary |
|
A worthwhile strategy for discovering useful designs for robot control systems is to 'reverse-engineer' biological control systems. In this project we focused on centralised structures in the vertebrate brain that form a critical part of the 'layered neural architecture' that controls behaviour. Computational models of two key integrative brain systemsthe 'reticular formation' and the 'basal ganglia'and of their inter-relationships, were developed and tested both in simulation and as parts of control systems for mobile robots. Both systems were shown to be capable of solving the action selection problemdeciding what 'to do next'but in different ways and with different strengths and weaknesses. We proposed that in the brain these systems form a partial hierarchy in which the basal ganglia selects actions and the reticular formation serves to co-ordinate sub-actions. In developing the model of the reticular formation we were able to demonstrate that the neurons in this part of the brain form a 'small world' networkan indication that these circuits possess interesting structural properties also seen in other real-world complex systems such as social networks and food webs. This is the first time a neural network in the brain has been shown to have this property. Finally, in developing robotic models of these brain systems we also generated new insights relevant to neuropsychology, in particular, in relation to the behavioural effects of the neuromodulator dopamine on behavior selection and switching.
|
| Further Information: |
http://www.shef.ac.uk/~abrg |
| Organisation Website: |
http://www.shef.ac.uk |
|
|