Planning of autonomous vehicle movement based on the available global map

Tema rada je istraživanje i razvoj algoritma za planiranje putanje autonomnog vozila. Predstavljena su neka dosadašnja rješenja te je definirano razvojno okruženje koje je korišteno za razvoj vlastitog rješenja. Za razvoj algoritma korišten je Udacityev simulator koji simulira uvjete na brzoj cesti s tri trake. Algoritam i simulator dvosmjernom komunikacijom razmjenjuju podatke, na način da simulator predaje senzorske podatke algoritmu koji zauzvrat vraća samo koordinate vozila koje mora posjetiti u narednom vremenu. Algoritam prvo učitava dane podatke, zatim vrši transformaciju u Frenetov koordinatni sustav radi lakšeg i preciznijeg proračuna. S Frenetovim koordinatama se računaju parametri ego vozila i moguća stanja koja može zauzeti na cesti. Potom se definiraju putanje za svako od stanja i ocjenjuju s obzirom na određene kriterije i pripadne težine. Nakon definiranje konačne putanje vozila, ona se mora transformirati natrag u Kartezijev koordinatni sustav, kako bi se predala simulatoru. Evaluacijom rješenja zaključeno je kako algoritam radi kako je i zamišljeno u uvjetima simulatora, ali to nije zadovoljavajuće za stvarne uvjete na cesti koji su nepredvidivi. Simulator nije realan prikaz uvjeta na cesti budući da simulira samo brzu cestu s ostalim vozilima, bez prometnih znakova ili promjene ograničenja. Što čini i same testove nerealnim prikazom rada algoritma. Implementacija algoritma na Alpha ploču nije osposobljena zato što nije ostvarena dvosmjerna komunikacija podataka između simulatora i ploče.The theme of the paper is the research and development of an algorithm for planning the path of an autonomous vehicle. The most important solutions so far are presented, and a development environment is defined in which the algorithm is made. For the development of the algorithm Udacity's simulator is chosen as it simulates the conditions on a fast three-track road. The algorithm and simulator communicate with the two-way communication; the simulator handles the sensor data to an algorithm that in return returns only the coordinates that the vehicle will pass in its trajectory. The algorithm loads the given data, and then transforms it into a Frenets coordinate system for easier and more precise calculation. Frenets coordinates calculate the parameters of the ego vehicle and the possible manners that can be taken on the road. Then, the paths for each of the states are defined and evaluated with respect to certain criteria and multiplied with respective weights. After defining the final path of the vehicle, it must be transformed back into the Cartesian coordinate system so that the simulator can handle the data. By evaluating the solution, it was concluded that the algorithm works as intended by the simulator, but this is not satisfactory for realistic road conditions that are unpredictable. The simulator is not a realistic view of road conditions since it only simulates a fast road with other vehicles, with no traffic signs or changes in constraints what makes the tests themselves unrealistic algorithm performance. The algorithm implementation on the Alpha board failed despite numerous attempts to enable communication between the simulator located on computer and the Alpha board via Ethernet