Browsing by Author "Tsourdos, Antonios"
Now showing 1 - 4 of 4
Results Per Page
Sort Options
Item Open Access A review of state-of-the-art 6D pose estimation and its applications in space operations(CEAS, 2024-06-11) Singh, Siddharth; Shin, Hyo-sang; Tsourdos, Antonios; Felicetti, LeonardIncrease in autonomous systems now requires for these systems to work in close proximity of other objects in their environments, with many tasks that need to be done on environment objects for eg., assembly, transportation, rendezvous, docking, or to avoid them like collision detections/avoidance, path planning etc. In this literature review we discuss machine learning based algorithms that solve the first step of vision-based autonomous systems i.e., vision based pose estimation. This paper presents a critical review in advancements of 6D pose estimation using both 2D and 3D input data, and compare how they deal with the challenges shared by the computer-vision based localisation problem. We also look over algorithms with their applications in space based tasks like in-orbit docking, rendezvous and the challenges that come with space-vision applications. To conclude the review we also highlight niche problems and possible avenues for future research.Item Open Access Disaster area coverage optimisation using reinforcement learning(IEEE, 2024-06-19) Gruffeille, Ciaran; Perrusquía, Adolfo; Tsourdos, Antonios; Guo, WeisiSearch and rescue applications using unmanned aerial vehicles (UAVs) also known as drones are becoming a research topic of interest to industry and academia due to its high impact in the ecosystem and people. Exploration of the disaster area is a crucial element in search and rescue to identify the zones that require immediate assistance or with high hazard probability. This paper aims to contribute in the coverage optimisation of a disaster area using drones. We focus on a flood disaster scenario as case of study. The proposed approach consists in two main parts: i) a Siamese-net is used to identify flooded buildings in satellite images, and ii) the points of interest are converted into a suitable maze environment that subsequently is used by any reinforcement learning (RL) architecture for area coverage optimisation. Here, the goal of the RL architecture is to ensure that the complete environment is covered by the drone by optimizing time and previously visited zones. Experiments are conducted to show the benefits and challenges of the current approach.Item Open Access RGANFormer: Relativistic Generative Adversarial Transformer for time-series signal forecasting on intelligent vehicles(IEEE, 2024-07-15) Xing, Yang; Kong, Xiangqi; Tsourdos, AntoniosTime-series modelling (TSM) is a critical task for intelligent vehicles (IVs), covering areas like fault detection, health monitoring, and inference of road user intentions. In this study, we present a novel TSM approach for enhancing the accuracy of multi-variate signal forecasting in intelligent vehicles. Our method leverages advanced Transformer networks within a relativistic generative adversarial network (RGAN) training framework. The RGAN training framework efficiently improves the accuracy of vehicle states forecasting for IV, demonstrating effective learning of long-time dependencies for more accurate predictions over extended sequences. Additionally, we introduce a high-dimensional extension (HDE) built-in block for the time-series Transformer to explore the impact of higher-dimensional features on representing long-term sequences. The experimental data is collected from a real-world electric vehicle testing bed. We evaluate the proposed RGANFormer framework and the HDE block on two popular time-series models, namely, Autoformer and FiLM. The results demonstrate that the RGANFormer, along with the built-in HDE block, significantly enhances long-term sequential forecasting accuracy for both multivariate and univariate tasks.Item Open Access Swarm decoys deployment for missile deceive using multi-agent reinforcement learning(IEEE, 2024-06-19) Bildik, Enver; Tsourdos, Antonios; Perrusquía, Adolfo; Inalhan, GokhanThe development of novel radar seeker technologies has improved the hit-to-kill capability of missiles. This is particularly worrying in safety and security domains that need the design of appropriate countermeasures against adversarial missiles to ensure protection of naval facilities. This paper aims to contribute in these domains by developing an artificial intelligence (AI) based decoy deployment system capable of deceiving the missile threat. Here, a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm is developed to maximise the distance between the target and the missile by learning the optimal/near optimal route planning of the six decoys to reach the global mission. As case study, the deployment of six decoys from the top deck of the main platform is assumed. The decoys are launched from the platform at the initial phase of the mission, and they establish a leader-follower formation that enhances the signal strength of the swarm decoys. The reward function is designed to guarantee a triangular formation configuration for swarm decoys. The reported results show that the proposed approach is capable to deceive the missile threat and has the potential to be integrated in current naval platforms.