COBEM99

SR21 Redes Neurais, Algoritmos Genéticos e Lógicas Nebulosa/Neural

Title:

GENERATION OF PLANE TRAJECTORY FOR MANIPULATOR RR TYPE USING NEURAL NETWORKS

Summary :

THE TRAJECTORY PLANNING IS AN IMPORTANT FACTOR IN A MANIPULATOR PRECISION. SO SMALLER DEVIATION OF POSITION AND ORIENTATION OF MANIPULATOR, RELATED WITH A KNOWN PATH, THAN BIGGER WILL BE THE MOVEMENT PRECISION. HOWEVER, SOME FACTORS HAVE AN INFLUENCE IN THE EXECUTION OF THE PATH. GENERALLY THESE FACTORS ARE TREATED EITHER BY NON-SYSTEMIC MODES, OR BY THE MATHEMATICAL MODEL IMPROVEMENT OR BY THE IMPLEMENTATION OF ELABORATED CONTROL TECHNIQUES, OR BY TASK CONSTRAINTS. THUS, THE USE OF NEURAL NETWORKS, ACTING AS A FILTER THAT IS ABLE TO SELF TUNNING WITH THE OPERATING MECHANISMS MOVEMENTS, CAN BE A WAY TO SIMPLIFY THESE PROBLEMS. IN THIS PAPER A NEURAL NETWORKS TYPE MULTI LAWYER PERCEPTRON HAS BEEN USED TO SIMULATE THE ROBOT SCARA TYPE MOVEMENTS IN A BIDIMENSIONAL PLANE, ACTING AS FOLLOWER OF KNOWN PATHS. THE WORKSPACE AND THE ANGULAR POSITION OF THE LINKS ARE MAPPED BY NEURAL NETWORKS THAT WILL RESPONSIBLE FOR THE MANIPULATOR MOVEMENT GENERATION. IT PRESENTS SOME SIMULATIONS OF MOVEMENTS THAT SHOWS THE NET PERFORMANCE IN THE BOUNDARY OF WORKSPACE AREA AND IN SOME CONFIGURATIONS THAT THE TRADITIONAL GEOMETRIC MODELS USUALLY PRESENTS SINGULARITY. KEY-WORDS: TRAJECTORY GENERATION, NEURAL NETWORKS, WORKSPACE.

Author :

Bezerra, Carlos Andre Dias
Mendeleck, André 0
Zampieri, Douglas Eduardo

Paper View :