Project Type: STREP
The aim of the SPARK Project is to develop new sensing-perceiving-moving artefacts inspired by the basic principles of living systems and based on the concept of "self-organization". Sensors will be treated as devices processing signals distributed in space and also showing non linear time dynamics. Perception will be studied as a result of a spatio-temporal pattern forming process, determined by information deriving from sensors and will directly influence the particular associated motor behaviour. The whole methodology will be implemented in a new architecture, a Spatial-temporal array computer based structure (SPARC), providing a new paradigm for active perception based on principles borrowed from psychology, synergetics, artificial intelligence and nonlinear dynamical systems theory.
The technical objective will be a moving artefact that will actively interact with the environment. It will integrate the sensor stimuli, will create an iconic, abstract and concise representation of the environment under the form of a dynamically emergent pattern in a SPARC based architecture and will generate a sequence of proper motor actions to reach a pre-specified target.
Start time : September 2004
End-time : August 2007
A) Representation and unification of Biological sensory systems under the framework of Spatial-temporal array computing (SPARC) architectures.
B) Introduction of a new paradigm for action-oriented perception, based on SPARC.
C) Implementation of the paradigm into a software/hardware environment with application to robot guidance.
- Insight into the various relevant types of sensing systems in animals (visual, hearing, tactile), able to provide a spatial-temporal simplified dynamic representation of the environment, useful for action purposes.
- Formalisation of the problem of perception as a pattern formation process in a spatio-temporal dynamical system.
- Derivation of spatial-temporal dynamical systems models, based on the SPARC structure, implementing the dynamic patterns described in the previous points.
- Implementation of the dynamical models into a simulation environment and design of some virtual experiments, using this software tool.
- Design of a unifying framework and methodology to extract information from the sensor models and to gain an abstract, action-oriented representation of the environment for decision making.
- Design of a robotic structure to act as a test bed for the methodology developed.
- Design of a mixed analogue/digital circuit environment to realise the methodology mathematically modelled at the previous points.
- Development of circuits paradigms and integration between the electronic and the mechanical part to build a real integrated sensing-perceiving-moving system. Construction of the robotic structure.