Posnemanje dinamskega modela mobilnega robota z uporabo LSTM nevronske mreže

Urh, Nino (author), Vrabič, Rok (mentor)

V magistrski nalogi je obravnavano posnemanje dinamskega modela mobilnega robota z diferencialnim pogonom z uporabo nevronske mreže z dolgim kratkoročnim spominom (LSTM). LSTM mreže imajo zmožnost analizirati pretekle dogodke, na katerih ustrezno napovedo kratkoročno prihodnost. Preverili bomo, ali mreža lahko naredi povezavo med vhodnimi in izhodnimi veličinami dinamskega modela. Najprej je predstavljena teorija o strojnem učenju in umetnih nevronskih mrežah ter delovanju fizikalnega modela robota. V praktičnem delu je opisana metoda generiranja podatkov in preizkusi različnih struktur LSTM mreže. Rezultati prikažejo, da umetna LSTM mreža dobro posnema dinamiko robota.

• Language: Slovenian
• Keywords: nevronske mreže, LSTM, dinamski model, robotika, diferencialni pogon, simulacija
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FS - Faculty of Mechanical Engineering
• Place of publishing: Ljubljana
• Publisher: [N. Urh]
• Year: 2020
• Number of pages: XXII, 75 str.
• PID: 20.500.12556/RUL-119357
• UDC: 004.85:007.52(043.2)
• COBISS.SI-ID: 28719619
• Publication date in RUL: 08.09.2020

Secondary language

• Language: English
• Title: Simulating a dynamic model with differential drive using LSTM neural network

Abstract

The masters thesis deals with imitation of a dynamic model of a differential-powered robot with an LSTM artificial neural network. Long-short term memory network (LSTM) have the ability to predict short term future events based on the past time-varied data. We will test the LSTM network's ability to make a corealtion between input and output variables of dynamic model. The theory of machine learning and artificial neural networks and the operation of a physical model of a robot are presented. The practical part describes the method of generating data and testing different structures of the LSTM network. The results show that it is possible to use the LSTM for replacement of the physical model without proper knowledge of the physical background.

• Keywords: neural networks, LSTM, dynamic model, robotics, differential drive, simulation