Optimizacija krmiljenja robotskega sledilca črti z metodo gradientnega spusta

Krasnik, Marko (author), Vrabič, Rok (mentor)

Zaključno delo obravnava uporabnost metode gradientnega spusta za optimizacijo krmilnih parametrov robotskega sledilca črti. Za preizkušanje je bil potreben program, naložen na mikrokrmilnik robotskega sledilca, ki poleg krmiljenja robota z uporabo gradientnega spusta išče optimalne vrednosti krmilnih parametrov. Preizkusi so potekali na testni progi, optimalne vrednosti krmilnih parametrov pa so izbrane tako, da je čas prevoza proge čim krajši. Izdelane so bile tri različice programa z manjšimi spremembani, z vsako od njih pa je bilo opravljenih več preizkusov. Z analizo rezultatov je bilo ugotovljeno, da je gradientni spust primerna metoda za optimizacijo krmilnih parametrov, saj je bil čas prevoza proge ob koncu preizkusov precej krajši. Za čim boljšo izvedbo optimizacije je pomembna tudi pravilna izbira kriterijske funkcije ter poznavanje, kako krmilni parametri vplivajo drug na drugega.

• Language: Slovenian
• Keywords: robotski sledilci črt, PID krmiljenje, gradientni spust, optimizacija, strojno učenje
• Work type: Final paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FS - Faculty of Mechanical Engineering
• Place of publishing: Ljubljana
• Publisher: [M. Krasnik]
• Year: 2019
• Number of pages: IX, 67 f.
• PID: 20.500.12556/RUL-110170
• UDC: 007.52:681.5:004.83(043.2)
• COBISS.SI-ID: 16931355
• Publication date in RUL: 12.09.2019

Secondary language

• Language: English
• Title: Optimisation of control parameters for a line follower robot using gradient descent

Abstract

This thesis deals with using the gradient descent method for optimizing control parameters of a line following robot. The robot's microcontroller was programmed to operate its control algorithm, as well as search for optimal values of the control parameters using gradient descent. Testing was done on a test track, optimal control parameters were chosen based on the fastest lap-times. Code for the microcontroller was witten in three separate versions with slight variations. Multiple tests on the track were done with each version. After analyzing our data, we can determine that gradient descent is an effective method for optimizing control parameters of a line following robot, since lap-times after optimisation have improved significantly. We have also concluded that choosing good crteria (in this case lap-time) and knowing how control parameters affect each other is very important for effective optimisation.

• Keywords: line following robots, PID control, gradient descent, optimisation, machine learning