Abstract
Indoor localization of robots is crucial for enabling effective navigation and path planning. Ultra-wideband (UWB) technology is gaining popularity due to its low cost and high accuracy. However, environmental obstructions often lead to Non-Line-Of-Sight (NLOS) signal propagation, which significantly impacts ranging and localization accuracy. Existing NLOS mitigation approaches typically discard identified NLOS measurements or utilize Channel Impulse Response (CIR), which may not be accessible from off-the-shelf UWB devices without directly extracting data from the UWB chip. To address these challenges, we propose the Compensated-Ranging Weighted Least Square (CR-WLS) localization approach, which mitigates the effects of NLOS ranging measurements without discarding them. Our approach focuses solely on the data available from off-the-shelf UWB devices without any extraction steps to obtain CIR data. The proposed method incorporates a Neural Network (NN) model trained on ranging and Received Signal Strength (RSS) data. The NN model outputs a weight and a compensation ratio for each ranging measurement. This weight and ratio are used to improve the localization accuracy using a Weighted Least Square (WLS) strategy. To validate the effectiveness of our approach, we conducted experiments in three different indoor environments. Our results demonstrate that the proposed CR-WLS approach outperforms the conventional Least Square (LS) approach, which does not consider NLOS, by up to 77.44% in localization accuracy.