Path Finding Autonomous Car, Designed For Room Mapping

Wienike, Thore Svejgaard ; Pedersen, Kristian Borup

10. term

Elektronik og IT, Kandidatuddannelsen (Spec. Intelligent Autonomous Systems)

2007

Projektet udvikler en lille, autonom bil, der kan videokortlægge et ukendt rum og anslå rummets mål. Arbejdet omfatter tre hoveddele: (1) analyse af, hvordan man bedst måler og kortlægger et rum, (2) design og konstruktion af en let bil og et kamerasystem, og (3) udvikling af bilens styring (controller/regulator). Analysen fokuserer på at vælge en egnet robotplatform til kortlægning og på at udnytte et lille “spy cam” optimalt for at opnå et komplet videokort. Bilen er konstrueret til at veje under 500 g og til at kunne estimere rummets dimensioner. For at udnytte kameraets lodrette synsfelt er der tilføjet en vippe-mekanisme til kameraet. Der er designet to controllere: en modelbaseret og en ikke-modelbaseret. Begge bygger på klassisk reguleringsteori, men forskellige metoder er brugt: Root Locus-metoden til den modelbaserede og Ziegler-Nichols-indstilling til den ikke-modelbaserede. Ydelsen er sammenlignet, og den ikke-modelbaserede controller gav de bedste resultater i simuleringer. Den kan dog kun bruges i simulation, mens den modelbaserede kan implementeres i den fysiske bil.

This project develops a small autonomous car that can create a video map of an unknown room and estimate the room’s dimensions. The work has three parts: (1) analyzing how to measure and map a room effectively, (2) designing and building a lightweight car and camera setup, and (3) designing the car’s controller. The analysis considers which robot platform is best for mapping and how to use a small “spy cam” to achieve a complete video map. The car is designed to weigh under 500 g and to estimate the room size. A tilting mechanism was added to the camera to make full use of its vertical field of view. Two controllers were designed: one model-based and one non-model-based. Both rely on classical control theory, but different methods were used: the Root Locus method for the model-based controller and Ziegler-Nichols tuning for the non-model-based controller. Their performance was compared. The non-model-based controller performed best in simulations, but it is only suitable for simulation, while the model-based controller can be implemented on the physical car.

[This abstract was generated with the help of AI]

Autonomous ; Mapping ; Autonom ; Kortlægning

Download
View record in AAU Student Projects