Machine Learning and UKF based Indoor Localisation using Ultra Wide Band (UWB) Sensors
Authors
Le, Pero ; Sreeram, Karanam Venkata ; Kanthappan, Rakesh Sundar
Term
4. term
Education
Publication year
2022
Submitted on
2022-06-01
Pages
115
Abstract
Accurate positioning is essential for autonomous robots to navigate safely indoors. In line-of-sight (LoS) conditions, common technologies such as GPS and Bluetooth often have uncertainties greater than 20 cm, whereas Ultra Wide Band (UWB) can achieve less than 8 cm. The problem becomes harder in non-line-of-sight (NLoS), where signals are reflected and attenuated and measurements become less reliable. This project explores tracking an indoor autonomous robot using UWB together with an Unscented Kalman Filter (UKF) to reduce NLoS effects. The robot, called “Deepcar” from SMPL Robotics, relies on UWB range measurements that the UKF fuses with a motion model to produce a robust position estimate even when measurements are noisy. To distinguish LoS from NLoS, a machine learning classification model analyzes the power of the received impulse responses at the beacon (anchor). The model was developed using the TREK 1000 evaluation kit. In the proposed approach, people in the environment carrying UWB tags are treated as non-stationary beacons (moving reference points). During NLoS, their measurements are assigned variable weights within the UKF so that information from both fixed beacons and human-carried, moving beacons helps improve Deepcar’s position estimate.
Præcis lokalisering er afgørende for, at autonome robotter kan navigere sikkert indendørs. Når der er direkte sigtelinje (Line of sight, LoS) mellem sender og modtager, har almindelige teknologier som GPS og Bluetooth ofte en usikkerhed på over 20 cm, mens ultrabredbånd (Ultra Wide Band, UWB) kan komme under 8 cm. Udfordringen bliver større uden direkte sigtelinje (Non line of sight, NLoS), hvor radiosignaler reflekteres og dæmpes, og målinger bliver mindre pålidelige. Dette projekt undersøger sporing af en autonom robot i indendørs miljøer ved hjælp af UWB og et Unscented Kalman-filter (UKF) for at mindske effekten af NLoS. Robotten, kaldet “Deepcar” fra SMPL Robotics, bruger UWB-baserede afstandsmålinger, som UKF’et kombinerer med en bevægelsesmodel for at give et robust positionsestimat, selv når målingerne er støjende. For at skelne mellem LoS og NLoS anvendes en maskinlæringsbaseret klassifikationsmodel, der analyserer styrken af de modtagne impulsresponser ved beaconen (fyr/anker). Modellen er udviklet med TREK 1000-evalueringssættet. I den foreslåede metode inddrages personer i det indendørs miljø, som bærer UWB-tags, som ikke-stationære beacons (bevægelige referencepunkter). Under NLoS tildeles disse målinger variable vægte i UKF’et, så information fra både faste beacons og menneskebårne, bevægelige beacons kan bidrage til at forbedre Deepcars positionsestimat.
[This apstract has been rewritten with the help of AI based on the project's original abstract]
Keywords