EMORPH

Event- driven morphological computation for embodied systems

Coordinatore	FONDAZIONE ISTITUTO ITALIANO DI TECNOLOGIA    more  Organization address contact info Cognome: BARTOLOZZI, CHIARA Email: send email Telefono: -71781382 Fax: -7170788
Nazionalità Coordinatore	Italy [IT]
Totale costo	1˙631˙545 €
EC contributo	1˙250˙000 €
Programma	FP7-ICT Specific Programme "Cooperation": Information and communication technologies
Funding Scheme	CP
Anno di inizio	2009
Periodo (anno-mese-giorno)	2009-02-01   -   2012-01-31

 Partecipanti

#	participant	country	role	EC contrib. [€]
1	FONDAZIONE ISTITUTO ITALIANO DI TECNOLOGIA	IT	coordinator	0.00
2	AUSTRIAN RESEARCH CENTERS GMBH - ARC  Organization address address: Donau-City-Strasse 1 city: WIEN contact info	AT (WIEN)	participant	0.00
3	UNIVERSITA DEGLI STUDI DI GENOVA  Organization address address: VIA BALBI 5 city: GENOVA contact info	IT (GENOVA)	participant	0.00
4	UNIVERSITAET ZUERICH  Organization address address: RAEMISTRASSE 71 city: ZURICH postcode: 8006 contact info	CH (ZURICH)	participant	0.00

Mappa

 Word cloud

Esplora la "nuvola delle parole (Word Cloud) per avere un'idea di massima del progetto.

 Obiettivo del progetto (Objective)

The goal of the eMorph project is to design asynchronous vision sensors with non-uniform morphology, using analog VLSI neuromorphic circuits, and to develop a supporting data-driven asynchronous computational paradigm for machine-vision that is radically different from conventional image processing. The mainstream computational paradigm in embodied intelligence is digital and it is clear that conventional digital systems have difficulties in performing robustly even in the most mundane tasks of perception. They require vast amounts of resources to extract relevant information, but still fail to produce appropriate responses for interacting with the real-world in real time. In addition, in sensory perception tasks, the data acquired from the sensors is typically noisy and ambiguous. 'Frame-based' time sampling and quantization artifacts present in conventional sensors are particularly problematic for robust and reliable performance.

The situation is clearly different in biological systems. In particular, biological neural systems vastly outperform conventional digital machines in almost all aspects of sensory perception tasks. Despite its dramatic progress, information technology has not yet been able to deliver artificial systems that can compare with biology. There are limitations both at the technological level, and at the theoretical/computational level. Analog computation - free from the limits of sampling - provides a solution. Analog devices are fast, as time constants are in the range of the rising time of the transistor currents. Event-driven computation intrinsically adapts the sensor response to the time constants of the real world. The sensor response is automatically regulated to match the incoming signal range, and so is robust. Moreover as only important events are coded, they are also efficient.

The eMorph project thus aims to design novel, data-driven, biologically inspired, analog sensory devices while also developing new asynchronous event-driven computational paradigms for them. eMorph aims to adapt the computational engine of the cognitive system (its morphology with respect to computation) to the dynamics of the real world rather than furiously sample the physical sensory signals in an attempt to obtain adequate bandwidth. Structure and morphology will be matched to the requirements of the robot's body and its application domain with testing to be carried out on the advanced humanoid robotic platform, iCub (project RobotCub, http://www.robotcub.org). The project will assemble a small but focused team of researchers from European leading institutions with well balanced complementary skills around these common goals.