Uche Onyekpe

With the development of autonomous vehicles, localization and mapping technologies have become crucial to equip the vehicle with the appropriate knowledge for its operation. In this paper, we extend our previous work by prepossessing a localization and mapping architecture for autonomous vehicles that do not rely on GPS, particularly in environments such as tunnels, under bridges, urban canyons, and dense tree canopies. The proposed approach is of two parts. Firstly, a K-means algorithm is employed to extract features from LiDAR scenes to create a local map of each scan. Then, we concatenate the local maps to create a global map of the environment and facilitate data association between frames. Secondly, the main localization task is performed by an adaptive particle filter that works in four steps: (a) generation of particles around an initial state (provided by the GPS); (b) updating the particle positions by providing the motion (translation and rotation) of the vehicle using an inertial measurement device; (c) selection of the best candidate particles by observing at each timestamp the match rate (also called particle weight) of the local map (with the real-time distances to the objects) and the distances of the particles to the corresponding chunks of the global map; (d) averaging the selected particles to derive the estimated position, and, finally, using a resampling method on the particles to ensure the reliability of the position estimation. The performance of the newly proposed technique is investigated on different sequences of the Kitti and Pandaset raw data with different environmental setups, weather conditions, and seasonal changes. The obtained results validate the performance of the proposed approach in terms of speed and representativeness of the feature extraction for real-time localization in comparison with other state-of-the-art methods.

This paper proposes a transfer learning approach to recalibrate our previously developed Wheel Odometry Neural Network (WhONet) for vehicle positioning in environments where Global Navigation Satellite Systems (GNSS) are unavailable. The WhONet has been shown to possess the capability to learn the uncertainties in the wheel speed measurements needed for correction and accurate positioning of vehicles. These uncertainties may be manifested as tyre pressure changes from driving on muddy and uneven terrains or wheel slips. However, a common cause for concern for data-driven approaches, such as the WhONet model, is usually the inability to generalise the models to a new vehicle. In scenarios where machine learning models are trained in a specific domain but deployed in another domain, the model's performance degrades. In real-life scenarios, several factors are influential to this degradation, from changes to the dynamics of the vehicle to new pattern distributions of the sensor's noise, and bias will make the test sensor data vary from training data. Therefore, the challenge is to explore techniques that allow the trained machine learning models to spontaneously adjust to new vehicle domains. As such, we propose the Recalibrated-Wheel Odometry neural Network (R-WhONet), that adapts the WhONet model from its source domain (a vehicle and environment on which the model is initially trained) to the target domain (a new vehicle on which the trained model is to be deployed). Through a performance evaluation on several GNSS outage scenarios - short-term complex driving scenarios, and on longer-term GNSS outage scenarios. We demonstrate that a model trained in the source domain does not generalise well to a new vehicle in the target domain. However, we show that our new proposed framework improves the generalisation of the WhONet model to new vehicles in the target domains by up to 32%.