P1. Develop an architecture for Cognitive Agents, including the following sub-systems:
S2. Attention Control
S3. Goals-oriented Computational Model
S5. Goal Generation system
S6. Concept System
S7. Reasoning system
S8. Action Generation system
P2. Develop further the Abstraction Mechanism to produce concept hierarchies on top of basic concepts.
P3. Implement the architecture in an embodied robotic agent, which will be tested in a ‘Patrolling, Incident Detection & Tracking Scenario’ in an outdoors environment.
The project’s duration is 36 months. The development of the core components will follow an order from lower level sub-systems to the highest. Figure 1 shows the timeline. Similarly coloured components are closely related.
Figure 1: Project Roadmap
M1. Perception Architecture (M06)
M2. Concept System Architecture (M12)
M3. Abstraction System Architecture (M18)
M4. Mobile platform commissioned (M16)
M5. Initial GNOSYS architecture implemented on mobile platform (M24)
M6. Final GNOSYS architecture implemented on mobile platform (M32)
M7. Test-bed and evaluation criteria for architecture assessment (M21)
I1. Exhibit ‘instinctive’ behaviours and act on them (curiosity, survival).
I2. Exhibit the generation of alternative virtual plans and selection of realisable plan biased through previous experience.
I3. Detection of novelties.
I4. Initiation of learning behaviours for novel environment.
I5. Creation of automated processing modes and integration with attentive responses. Automated modes represent reactive responses of learned action-sequences for given stimulus. Automated modes are monitored by attention system for response failures.
I6. Multi-modal perceptual fusion
I7. Overall fused decision (action generation) generation from competing modalities.
I8. Concept formation using variable dimension feature vectors.
I9. Abstraction of concepts and development of concept hierarchies during operation.
I10. Creation of goals and biasing of attention.
I11. Development of value systems, emotional states and implementation of internal drives.
I12. Development of sensory and motor attention per modality.
I13. Development of computational model which prioritises responses based on global attention competition, internal drives/value systems and projected agent states.
I14. Integration of user-level goals with own goals. User-level goals change priority depending on the observed external and internal states and other competing goals.