Pneumatic Manipulator with Neural Network Control
Can the results of task solution be improved by means of neural net \( {\varphi }^{ - 1} \) implementation? Traditional methods gives ideal solution for ideal arm link control model. Real arm link and it’s control can slightly differ from ideal model. Dur
- PDF / 956,605 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 56 Downloads / 255 Views
Keldysh Institute of Applied Mathematics of RAS, Moscow, Russia {aliseychik,i.orlov}@keldysh.ru, {vlpavl,smolin}@keldysh.ru 2 Department of Mechanics and Control of Machines, Institute for Machine Science Named After A.A.Blagonravov of RAS, Moscow, Russia {aliseychik,i.orlov}@imash.ru 3 Lomonosov Moscow State University, Moscow, Russia [email protected] 4 Russian State University for Humanities, Moscow, Russia [email protected]
Abstract. Can the results of task solution be improved by means of neural net u1 implementation? Traditional methods gives ideal solution for ideal arm link control model. Real arm link and it’s control can slightly differ from ideal model. During traditional control installation and tuning only few control parameters could be adjusted, while the function shape and direct proportionality control actions to pulse-width modulation (PWM) parameters are unalterable. Neural nets are representing transformation function in tabular form and are capable for functions form fine tuning during adjustments to real control conditions. The reported study was funded by RFBR, according to the research project No. 16-38-60201 mol_a_dk. Keywords: Manipulator
Control system Neural network
1 Introduction Manipulator ManGo with SCARA-like kinematics with pneumatic actuators considered in this paper. The original concept of position control based on neural network is proposed as a pneumatic actuator control system. Dynamic model and control system implemented in Matlab Simulink. The low-level control system is implemented on a microcontroller STM32F4 Discovery. Games “Go” and “Gomoku” selected as experimental tasks. Machine vision system for the recognition of the board and the game situation are implemented on the Android OS using OpenCV [1]. Task of recognising go board and stones is not new, as are methods which were used in process of solving it. Most of mathematical algorithms as FLANN and Hit-or-Miss algorithms are well known and commonly used, especially in computer vision. The main goal was to © Springer International Publishing Switzerland 2016 L. Cheng et al. (Eds.): ISNN 2016, LNCS 9719, pp. 292–301, 2016. DOI: 10.1007/978-3-319-40663-3_34
Pneumatic Manipulator with Neural Network Control
293
achieve a higher level of accuracy of recognition than of available solutions to the task. The robot was piloting, allowing to draw conclusions about the effectiveness of created software and hardware solutions to control manipulators with pneumatic actuators.
2 Manipulator ManGo ManGo manipulator (Fig. 1) has a SCARA-like kinematic [2], that mostly suits object manipulation tasks on a plane, including desk games. The first steps in kinematic analysis were carried out during the creation of robots design in CAD soft complex, the pneumatics of Italian company Pneumax was used as the executing motor.
Fig. 1. ManGo robot
Optimal lengths of partsand attachment points of pneumatics were calculated to cover workspace of 500 × 500 cm size, which is enough to work with almost every knowledge-based logical d
Data Loading...
