Teaching

The usual sensing modalities that are available to autonomous vehicles i.e. vision, radar and lidar have their limitations. For example, none of them is very suitable for estimating the length of a leading vehicle. Additional sensing modalities, such as wind flow sensing, might have the potential to meaningfully complement the usual sensing modalities in autonomous driving.You will build upon the innovative work done on flow-sensing applications in autonomous driving two previous theses at ADL There are multiple directions that can be investigated. For example, (i) working in simulations and proposing new applications of flow sensing in autonomous driving, (ii) investigating new feature extraction and classification/regression methods etc. or (iii) expanding out of simulations with physical validation of flow sensing in autonomous driving.

Autonomous vehicles, like other robots, need to localize themselves in order to navigate. While satellite navigation systems (GNSS) such as GPS can provide such vehicles with localization information, the GNSS information might not always be available. One robust technique for vehicles to localize is using particle filter, given a map of the environment. Recently, deep-learning approaches to map-based localization have also been proposed in the literature. During the project, you will investigate lidar sensors for map-based localization. You can start by experimenting with lidar data acquired using the ADL Lexus vehicle and then move to real-time implementation on-board the vehicle, within our Autoware Mini software stack. 

Autonomous vehicles research has progressed rapidly during the last decade, but challenges remain in their wide-spread deployment. Modular as well as end-to-end approaches to vehicle autonomy, alike, suffer from the limitation of not being able to cover edge cases that arise in everyday traffic. At the same time, it is also believed that more and more infrastructural support and interconnectivity between vehicles will be available to autonomous vehicles of the future over time. In this project, you will have the possibility to propose an autonomous-driving system which can get as much information from infrastructure and other road agents (including, but not limited to, other vehicles, pedestrians, bicycles etc.) as needed to be fail safe in all usual or edge situations. After a thorough literature review of the open challenges in vehicle autonomy you will investigate methods for meeting those challenges using connectivity with infrastructure and road agents. You will then validate the proposed system in simulation.

Autonomous vehicles, like other robots, need to localize themselves in order to navigate. While satellite navigation systems (GNSS) such as GPS can provide such vehicles with localization information, the GNSS information might not always be available. This is where cross-view geolocalization comes in. It refers to performing localization using vehicle’s sensors (such as a camera) and using information from a completely different modality such as satellite images as a map of the environment that you are trying to localize the vehicle in. Recently, many works have been published on cross-view geolocalization in urban environments such as cities. In this project, you will replicate one of the above mentioned works, and study it for localization in non-urban environments i.e. outside the cities. The challenge here lies in the fact that urban environments are more feature rich (residential houses, other buildings etc.) in comparison to non-urban environments.

Autonomous vehicles, like other robots, need to localize themselves in order to navigate. While satellite navigation systems (GNSS) such as GPS can provide such vehicles with localization information, the GNSS information might not always be available. This is where cross-view geolocalization comes in. It refers to performing localization using vehicle’s sensors (such as a camera) and using information from a completely different modality such as satellite images as a map of the environment that you are trying to localize the vehicle in. Recently, many works have been published on cross-view geolocalization in urban environments such as cities. In this project, you will implement at least three methods for cross-view geolocalization on a single dataset, in order to quantitatively compare the methods. Depending on the exact methods from the literature that you choose to compare, you are welcome to use the code from the authors of the studies in case it’s publicly available. You are encouraged to propose at least one improvement to one of the (compared) methods.  You will then incorporate one of the methods (preferably the method you proposed an improvement to) into our Autoware Mini software stack.

Developing a course (8 academic hours one-day seminar or a 3 ECTS course) teaching key aspects of data science in a practical way with self-driving toy cars. The particular emphasis is on aspects that become relevant when deploying a solution to the real world. The toy cars are equipped with a camera and a Raspberry Pi. The used machine learning solution imitates human driving, learning the function between the camera image and the command a human would give when seeing this image. As the task is fun and understandable, the failures are evident (car crashes or behaves weirdly), it is a very good medium to teach data science and machine learning. The course should cover key concepts that get mentioned in every data science, but rarely practically experienced by students, such as the following. (i) Garbage in, garbage out. If your training data is low quality, you get a bad model. Can be exemplified by using too low camera resolution for the task (garbage input) or bad driving examples (garbage training labels). (ii) Overfitting and finding non-causal relationships in the data, which later lead to bad generalization ability. Exemplified by visualizing the image areas that influence the decision the strongest (saliency maps). If top half of the image is not cut away, the model will learn to rely on lamp locations in the ceiling, not track walls for deciding when to turn.  (iii) Models struggle to extrapolate, generalize. Exemplified by models failing to drive if light conditions change. (iv) Test set metrics are not the final product, the model will be deployed in the real world and will find examples it can not deal with. Or will experience a shift in input distribution. Monitoring performance, characterizing failures and reacting to them is important. (v) Dataset management, iteratively building a better dataset (to fix the failures observed), and data cleaning are important. Exemplified by results before and after. (vi) Computational efficiency can be important in certain tasks. If the vehicle takes too few decisions per second it will crash. In other domains - if every query costs too much (on compute bill) or takes too long to compute, the product is not viable. The thesis should make use of known theories of developing educational materials, e.g. define learning objectives (what should the student know) and work backwards from there. In particular for a MSc thesis, the educational theory background is needed for a good mark. MSc thesis would also benefit from running the developed course once with test students. BSc thesis can get away without actual experimentation due to time constraints.

The Autonomous Driving lab owns a dozen DonkeyCar S1 minicars. These vehicles come equipped with a camera and inertial measurement unit, as well as Raspberry Pi. Your task is to localize the vehicle with the best possible accuracy within the third floor of Delta building (starting with only one wing or hallway, for example) using the available sensors. This means the high-level localization (which hallway I’m in) can be based on the proximity of wifi access points, while more precise localization inside a certain hallway can be based on lines detected in the camera image. The supervisors see particle filters as a possible approach to the task, as it allows combining multiple sources of information (north direction from IMU, hallway-precision from wifi, location within hallway by camera). However, the student should complete a thorough literature review on camera-based localization and another method may prove applicable. If the vehicle is capable of localizing itself within the third floor of Delta with sufficient precision, it can follow paths and complete any routes given to it, opening possibilities for further applications, such as delivery of items, security patrols etc.