Cooperative driving of a cargo trolley with wheeled mobile robots

Prevoz tovora se v logistiki in proizvodnji ponavadi opravlja z večjimi mobilnimi roboti, ki pa za infrastrukturo niso vedno najbolj primerni. So tudi težki, neokre- tni in zahtevajo veliko prostora, tudi ko ničesar ne prevažajo. V tem magistrskem delu je predstavljen algoritem za prevoz tovora z več manjšimi mobilnimi roboti, ki si tovor delijo. Manjši mobilni roboti so lahko lažji, hitrejši in bolj okretni, če je potrebno lahko učinkovito prevažajo tudi manjše tovore. V primeru večjega tovora pa se roboti s posebnim prijemalom pritrdijo na kolesa tovornega vozička, nase prevzamejo del teže tovora in voziček s tovorom odpeljejo na želeno mesto. Za prevoz tovora med točkami je bil razvit centraliziran algoritem, ki omogoča vodenje poljubnega števila robotov z diferencialnim pogonom. Algoritem je se- stavljen iz treh delov: sistema za izračun kinematike vožnje tovora, sistema za določanje orientacije robotov glede na tovor in dela, ki skrbi za vodenje robotov. Orientacija robotov glede na tovor se izračuna na podlagi treh metod, ki temeljijo na globalni legi robotov, odometriji robotov in kinematičnem modelu ter vizu- alnih markerjih. Za iskanje poti se uporablja algoritem A*, za sledenje poti pa prediktivni algoritem načrtovanja in sledenja trajektorije (Trajectory Rollout). Algoritem pretvori kontrolne ukaze za premik tovora iz algoritma za sledenje poti v kontrolne ukaze za posamezne mobilne robote. Implementacija algoritma temelji na okolju ROS (Robot Operating System). Algoritem je bil razvit in ovrednoten v simulacijskem okolju Gazebo. Nato se je prenesel in ovrednotil še na realnem sistemu – mobilnem robotu podjetja Epilog d.o.o. V simulaciji in na realnem sistemu so bili opravljeni eksperimenti z odprto- in zaprtozančnim vodenjem. Ustrezno delovanje algoritma se je določilo s tremi pokazatelji. Prvi pokazatelj je primerjava hitrostnega ukaza z dejansko hitrostjo tovora. Drugi pokazatelj so sile med roboti in tovorom, tretji pa bočno drsenje robotov. Algoritem je v vseh simuliranih in realnih eksperimentih deloval kar je pomenilo, da se je tovor premikal v skladu z želenimi ukazi brez bočnih drsenj robotov in brez prevelikih sil.Transport of load in production and logistics is usually done with bigger mobile robots that are not always the most suited for infrastructure. They are heavy, slow, and require lots of space even when not transporting anything. This mas- ter’s thesis presents the algorithm for transportation of load with multiple smaller mobile robots that share the load. Smaller robots are lighter, faster, and require less space. If needed they can more effectively transport smaller loads. In the case of bigger load robots use special gripper to attach themselves to the wheels of the transport cart and transfer some weight to themselves. A centralized algorithm was developed that is capable of controlling any num- ber of differential drive robots with the aim of transporting a load between points. The algorithm consists of three parts: a system for calculating kinematics of load transport, a system for determining the orientation of the robots with respect to load and robots control system. The orientation of robots with respect to load is calculated based on three methods that use the global pose of the robots, odom- etry of the robots and kinematic model, and visual markers. For path searching A* algorithm is used and for path following Trajectory Rollout algorithm is used. The algorithm is used to transform control orders for the load to controls for individual robots. The implementation of the algorithm is based on ROS (Robot Operating System) environment. The algorithm was developed and evaluated in Gazebo simulation environ- ment. It was then transferred and evaluated on a real system - on the mobile robot from company Epilog d.o.o. Open and closed-loop control experiments were done in simulation and on the real system. The proper working of the algorithm was determined using three indicators. First indicator in comparison of desired versus real velocity of the load. The second one is the forces between robots and the load. The third is sideways sliding of the robots. The algorithm worked in all simulated and real experiments which meant that the load was moving according to the commands without sideways sliding and without too big forces