COgnitiveSystems using Perception-Action Learning

SIXTH FRAMEWORK PROGRAMME
Information Society Technologies
IST-2003-2.3.2.4 Cognitive Systems
Project Number: IST-004176

COSPAL final deliverables

The Final Activity Report describes contributions to the research field, achievments and scientific impact of the COSPAL project.
The Final Demonstrator Report describes the COSPAL demonstrator. A short video sequence shows all modules interacting to control the robot.

Articles about the COSPAL project at the ICT Results website

Part 1: New breed of cognitive robot ... a puppy?
Part 2: Robotic minds think alike?

Article about the COSPAL project at ZDNet

A new kind of cognitive autonomous robots

Article about the COSPAL project at The Engineer

Early learning centre

Executive Summary

Artificial Cognitive Systems (ACS) will become a future key technology with high impact on economy, society, and daily life. Autonomous robots with perceptual capabilities can be applied in many areas such as production, public and private service, science, etc. The current state of research on ACS shows essential lacks concerning the design and architecture of the systems. In the COSPAL project, we try to overcome these limitations and to propose a new ACS architecture.

In the COSPAL architecture we combine techniques from the field of artificial intelligence (AI) for symbolic reasoning and learning of artificial neural networks (ANN) for association of percepts and states in a bidirectional way. We establish feedback loops through the continuous and the symbolic parts of the system, which allow perception-action feedback at several levels in the system. After an initial bootstrapping phase, incremental learning techniques are used to train the system simultaneously at different levels, allowing adaptation and exploration. We expect the COSPAL architecture to allow the design of systems that show to a large extent autonomous behavior.

Roadmap / Milestones (start date: 1/7-2004):

• Month 12: first results of partial aspects of the architecture
  • Continuous perception-action feedback and medium level learning
  • The bidirectional symbol-state interface and symbolic representations of dynamic states
• Month 21: first demonstrations
  • Implementation of essential system parts
  • Bootstrapping and incremental learning
• Month 34: final demonstrator
  • Symbolic reasoning and learning in the COSPAL architecture
  • Medium-level perception
  • Implementation of purpose in the system
  • Final implementations

Success indicators:

• Reports:
  • Conceptual and technical soundness of theory
  • Practical improvements documented by experiments and comparisons
• Prototypes:
  • Working implementations of modules
  • Working inter-module communication
• Demonstrators:
  • Documented and optimized implementations for all required modules
  • A successfully working demonstrator, i.e., the system learns to solve shape-sorter puzzles