Testing Nividia Drive for small cars in toy town
Organisatsiooni nimi
Autnonomous Driving Lab
Kokkuvõte
Nvidia has released a set of trained neural networks related to self-driving cars. These can be used as-is on whatever data you feed to them. These include models that detect objects (cars, signs, pedestrians), traffic lights, free area (drivable area), edges of the road etc.
(See: https://docs.nvidia.com/drive/archive/driveworks-3.0/index.html)
In this thesis, the student collects data in a toy town with a toy car (DonkeyCar) and applies these networks built for real-life data on the toy town data. Gautam has applied them to online datasets and can support setting up the pipeline. We measure qualitatively and wherever possible quantitatively the performance of these networks on toy town data. Any alterations to the toy town that are needed are possible in the spring semester (e.g. adding lanes/road edges by tape).
If these out-of-the-box solutions work for the toy data, they may be used in the future for making the toy cars avoid objects and stay on the road.
NB! This project is mainly for managing data, developing a pipeline to apply existing networks, computing and summarizing performance, selecting performance metrics. You do not train new neural networks here, just apply them.
(See: https://docs.nvidia.com/drive/archive/driveworks-3.0/index.html)
In this thesis, the student collects data in a toy town with a toy car (DonkeyCar) and applies these networks built for real-life data on the toy town data. Gautam has applied them to online datasets and can support setting up the pipeline. We measure qualitatively and wherever possible quantitatively the performance of these networks on toy town data. Any alterations to the toy town that are needed are possible in the spring semester (e.g. adding lanes/road edges by tape).
If these out-of-the-box solutions work for the toy data, they may be used in the future for making the toy cars avoid objects and stay on the road.
NB! This project is mainly for managing data, developing a pipeline to apply existing networks, computing and summarizing performance, selecting performance metrics. You do not train new neural networks here, just apply them.
Lõputöö kaitsmise aasta
2021-2022
Juhendaja
Ardi Tampuu & Gautam Kumar Jain
Suhtlemiskeel(ed)
inglise keel
Nõuded kandideerijale
Tase
Bakalaureus
Kandideerimise kontakt
Nimi
Ardi Tampuu
Tel
E-mail