Prediktivno vodenje kolesnih robotov v industrijskem okolju

RAVNIKAR, ROBI (author), Klančar, Gregor (mentor)

Zaradi vse večjega vključevanja ljudi v delovni prostor je narasla potreba po zanesljivem algoritmu, ki je sposoben detekcije in izogibanja ovir. V ta namen so se razvili prediktivni algoritmi vodenja, ki vnaprej predvidevajo trk z objektom in robota predhodno preusmerijo. V magistrski nalogi bom predstavil algoritem, ki rešuje problematiko vodenja mobilnih robotov v industrijskem okolju. Predstavljen algoritem se naslanja na bolj znan pristop dinamičnega okna in dinamičnega generiranja trajektorij ter skuša odpraviti njune pomanjkljivosti. Povečano je število možnih poti, med katerimi algoritem lahko izbira, ter njihova razvejanost. Dodal sem kompenzacijo časovnega zamika med zajemom lokalizacijskih podatkov ter njihovo uporabo v izračunu poti. Poleg osnovnih cen za odmik od cilja in poti sta na voljo še ceni za spremembo translatorne in kotne hitrosti, ki omogočata boljši nadzor nad dinamiko vozila. Algoritem sem testiral tako v simulacijskem okolju Gazebo kot v realnem okolju na robotu podjetja Epilog d. o. o. V simulacijskem okolju sem najprej preveril njegovo osnovno delovanje, nato pa še sposobnost izogibanja statičnim oviram. Na robotu pa sem opravil test gibanja iz točke A v B, izogibanje statični oviri ter izogibanje gibajoči se osebi. Rezultate vseh testov sem predstavil v obliki dveh grafov, ki prikazujeta razdaljo do končnega cilja ter napako sledenja globalni poti. V zaključku sem še opisal večje pomanjkljivosti algoritma ter predlagal možne izboljšave.

• Language: Slovenian
• Keywords: mobilna robotika, kolesni roboti, vodenje, prediktivno vodenje, generiranje trajektorij, dinamično okno, lokalno planiranje, planiranje poti, izogibanje oviram
• Work type: Master's thesis/paper
• Organization: FE - Faculty of Electrical Engineering
• Year: 2021
• PID: 20.500.12556/RUL-131087
• COBISS.SI-ID: 77907971
• Publication date in RUL: 22.09.2021

Secondary language

• Language: English
• Title: Predictive control of wheeled robots in an industrial environment

Abstract

Due to an ever increasing inclusion of people into the workspace of robots, the demand arose for a reliable algorithm that is capable of obstacle detection and avoidance. To fulfill that demand, a group of predictive control algorithms were developed that can predict collisions with objects and divert the robot in order to avoid them. The algorithm I present in this thesis is based on the more widely known dynamic window and trajectory rollout algorithms and tries to correct their flaws. The number of additional possible paths is increased and so are their branching options. I added a time delay compensation to compensate for the time difference between the capture of localization data and their usage in the algorithm. Besides the base cost parameters of distance to goal and path following error, two new were added. The costs of translational and angular velocity change, help increase the users control over the dynamics of the robot. I tested the algorithm both in the Gazebo simulation environment software and in a real environment with a mobile robot developed by the company Epilog d.o.o. The simulation was first used to test the basic functionalities of the algorithm and later its ability to avoid static obstacles. With the real robot, we performed tests that determined if it could travel from point A to B, avoid static obstacles and avoid a moving person. In the conclusion I describe all the major flaws of the algorithm and ways to improve it.

• Keywords: mobile robotics, wheeled robots, control, predictive control, trajectory rollout, dynamic window approach, local planning, path planning, obstacle avoidance