A MAN-MACHINE INTERFACE BASED ON A PETRI NET NEURAL CONTROL
FOR MOBILE ROBOTICS APPLIED TO HANDICAPPED PERSONS
Alexandre Abellard1, Patrick Abellard1,2, Philippe Grall1, Nicolas Razafindrakoto3 and Mohamed Ben Khelifa1
1 Laboratoire SIS, Université de Toulon, BP 20132, 83957 La Garde Cedex, France
2 CERTV, Université de Toulon, BP 20132, 83957 La Garde Cedex, France 3 Ecole Supérieure Polytechnique d’Antananarivo, BP 1500, 101 Antananarivo, Madagascar E-mail : {alexandre.abellard, abellard, ph.grall,khelifa}@univ-tln.fr
Tél : +33.(0)4.94.14.21.48
KEYWORDS
Petri Nets, neural networks, ship collision avoidance, visual collision detection, low-frequency vibrations for perturbation.
ABSTRACT
The research of solutions to the problem of hardware/software codesign, is a major task in the definition of a unique, structured and automatic methodology, providing the acceleration of design process and the dynamic evaluation of different compromises. Data Flow Petri Nets are an efficient solution to make it possible and the use of a hardware description language, allows their implementation on programmable chips. The example described in this paper deals with a wheelchair whose commands can adapt to the handicap. For some handicapped people, the use of a wheelchair can be difficult, due to weak physical capacities or cognitive troubles. So, the human-machine interface must be modular, configurable and easy to implement. It must bring reliability and use non specific material as often as possible. Therefore, the FRACAH project (Fauteuil Roulant A Commande Adaptée au Handicap) has been developed. Its lever is handled by an artificial neural network that records functional limitations of the hand, and then compensates them.
INTRODUCTION
The hardware/software codesign deals with the conception of systems including a material part with specific chips (ASIC, reconfigurable components…) and a software part, that is executed on an architecture based on standard or specific processors (microcontrollers, DSP...). It deals with specification, validation, and exploration of different possibilities for the design of a system in order to optimize the cost and/or the performance criteria (Dias 2000). The research of the solutions is a crucial task in the definition of a unique, structured and automatic methodology, as it enables us to define fundamental characteristics such as cost and temporal performances (Dehon et al. 1996).
The advantages of the integration of hardware and software design are the acceleration of the conception process, and the possibility to dynamically evaluate different possible compromises (Rompaey et al. 1986). For instance, in the case of DSP and specific circuits, the design process must suppress the existing gap between the heterogeneous functional specification and its implementation. At the system level, DSP need a combination between data flow and control flow models for their complete specification (Rubini 1997). The use of Data Flow Petri Nets is an efficient way to do this.
In fact, the algorithm to implement can be easily described in this manner, and simulated with any standard tool. From the simulation results, it is possible to write a VHDL program and to compile it for its implementation on programmable chips (Belhadj 1994).
DATA FLOW PETRI NETS
They have been defined by J. Alhmana (Alhmana 1983) and enable to efficiently define multiprocessors machine of data flow/control flow type, in which the research of performances implies to overcome various difficulties such as : bus sharing, memory access conflicts, inter-processor exchanges... This kind of architecture is structurally different, as it includes neither a central processor, nor a central memory. Data direct the operations. Many applications with various functional modes do exist (Abellard and Grall 2000).
Data Flow Petri Nets (DFPN) make the combination of data flow and control flow in one model. They are used as a tool of specification, simulation and data flow architectures validation (Abellard et al. 2001). This model, based on the notion of two-part places (operators Po and variables Pv) allows the modelling of different
.....