Current page: Members->List of Members->Murray Shanahan
 

Murray Shanahan
Imperial College London
Membership Number: 48
Address: Department of Electrical and Electronic Engineering, Exhibition Road, London SW7 2BT, United Kingdom
Email: m.shanahan@imperial.ac.uk
Phone: +44 (0)207 594 6307
Fax:
URL: http://www.iis.ee.ic.ac.uk/~mpsha/

Biographical Sketch

Imperial College London, Department of Electrical and Electronic Engineering

The Robotics and Artificial Intelligence research group has been very active in the areas of knowledge representation, high-level robot perception, and cognitive robotics. The group’s main contributors in this area are Murray Shanahan, Dave Randell, and Mark Witkowski. Much of this has had a series of UK EPSRC-funded projects (“Cognitive Robotics”, henceforth CR-I, “Cognitive Robotics II”, henceforth CR-II, and “Spatial Reasoning and Perception in a Humanoid Robot”, henceforth SRP).

Building on Shanahan’s earlier research on reasoning about action using the event calculus, and his prize-winning work on robot perception through abduction, the first of these projects (CR-I) demonstrated the feasibility of high-level robot control through logic programming. The architecture developed on the project combined sensor data assimilation, hierarchical planning, and reactivity in a single framework based on abductive logic programming.

In the follow-on project (CR-II), the emphasis was on scaling up the earlier results and applying them to richer sensory modalities, in particular to vision. With the appointment of Dave Randell, whose work on qualitative spatial reasoning is highly influential in the knowledge representation community, the group was in a unique position to explore the interface between robot perception and spatial reasoning. This led to original work on two fronts. First, a logic-based calculus for reasoning about spatial occlusion was formulated, along with methods for automated reasoning about the relations it deploys. Second, the earlier abductive account of sensor data interpretation was extended to handle top-down information flow through expectation. As in the preceding work, logical abduction was used to form a set of initial hypotheses to explain the low-level image data. But in the new approach, a deductive component subsequently computes the expectations of each competing hypothesis. The raw image is then re-consulted in order to check which expectations are fulfilled, resulting in the confirmation or disconfirmation of each hypothesis. A logic programming implementation was also developed that has demonstrated the computational feasibility of the technique. In the ongoing project (SRP), this work on top-down information flow is being further extended to accommodate active perception in the context of a humanoid robot that can nudge objects to see them from different angles. Again, this is facilitated by the top-down influence of a reasoning component, which can determine exactly which actions to perform in order to confirm or disconfirm an ongoing interpretive hypothesis. Concurrently, the computational issue has been pursued further, and the representational power of the event calculus has been reaffirmed.


Site generated on Friday, 06 January 2006