Animals for surviving have developed cognitive abilities allowing them an abstract representation of the environment. This internal representation (IR) may contain a huge amount of information concerning the evolution and interactions of the animal and its surroundings. The temporal information is needed for IRs of dynamic environments and is one of the most subtle points in its implementation as the information needed to generate the IR may eventually increase dramatically. Some recent studies have proposed the compaction of the spatiotemporal information into only space, leading to a stable structure suitable to be the base for complex cognitive processes in what has been called Compact Internal Representation (CIR). The Compact Internal Representation is especially suited to be implemented in autonomous robots as it provides global strategies for the interaction with real environments. This paper describes an FPGA implementation of a Causal Neural Network based on a modified FitzHugh-Nagumo neuron to generate a Compact Internal Representation of dynamic environments for roving robots, developed under the framework of SPARK and SPARK II European project, to avoid dynamic and static obstacles.
[1]
Valeri A. Makarov,et al.
Compact internal representation of dynamic situations: neural network implementing the causality principle
,
2010,
Biological Cybernetics.
[2]
Péter Szolgay,et al.
Configurable multi-layer CNN-UM emulator on FPGA
,
2002,
Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications.
[3]
P Arena,et al.
Image processing with cellular nonlinear networks implemented on field-programmable gate arrays for real-time applications in nuclear fusion.
,
2010,
The Review of scientific instruments.
[4]
Valeri A. Makarov,et al.
Compact internal representation as a protocognitive scheme for robots in dynamic environments
,
2011,
Microtechnologies.
[5]
José Manuel Ferrández,et al.
High Performance Implementation of an FPGA-Based Sequential DT-CNN
,
2007,
IWINAC.