Basic objectives of the project SCANDLE
The aims of the project will be achieved through the following scientific and technological objectives that fall within the four core areas of –
1) Architecture and hardware embodiment
- To develop an innovative distributed active/passive sound-processing system realized in real-time neuromorphic spike-based technology. The system will be capable of autonomous self-configuration and context-dependent organisation of perceptual representations of animate entities in the environment, derived both through sounds emitted by those entities, and from the sonar echoes resulting from sounds emitted by the system. Distributed elements of the system will communicate using wireless spike-based technology in order to arrive at global integrated representations of the detected animate entities.
2) Scene analysis and modelling in natural systems
- To extend current scientific knowledge regarding the parsing and perceptual organisation of multi-source scenes in natural systems through the construction of an internal model of the world and entities within it. Experiments in humans will be designed to probe the representational structures and processes necessary to support the creation and maintenance of auditory objects as well as the mechanisms which allow flexibility in human perceptual organisation. The neural correlates of perceptual organisation and context-dependent processing will be investigated in finer detail using an animal model, adopting similar experimental protocols to those used in human studies.
3) Computational modelling and software simulation
- To develop a theoretical understanding of the processes underlying autonomous perceptual organisation and pattern recognition, instantiated in the form of computational models. The computational models will be formulated at an abstract level in terms of interacting neural populations, as well as at the more detailed level of spiking neurons, and will have two important roles; firstly to verify theoretical ideas in relation to experimental results, and secondly to support the implementation of an autonomous neuromorphic system by modelling at some level of detail the neural basis for context-dependent processing in natural systems.
4) Environment and evaluation
- To develop the hardware and software infrastructure and computational components necessary for implementing an integrated acoustic scene analysis system capable of operating in a truly autonomous manner on low-energy hardware. The problem of situated cognition is hierarchically solved by obtaining answers to the questions: Is there anything interesting in the environment, where is it and what is it? To that end we will exploit a multitude of passive and active sensors, including continuous-wave acoustic micro-Doppler sonar embedded in a distributed network. The aim is to match or even exceed the cognitive capabilities of humans in the specified scenarios. We will develop and test the component subsystems necessary for building the proposed prototype system, explore the hardware implementation of novel algorithms and architectures, test the distributed control and sonar hardware, and evaluate the integrated system in a variety of real-world scenarios in comparison with human performance in similar situations.