UWB/IMU Integration with Adaptive Motion Constraints to Support UXO Mapping

A platform equipped with electromagnetic inference (EMI) sensor allows us to map underground areas and searching for unexploded ordnances (UXO). This mapping requires the precise navigation of the platform. In this paper, we use UWB/IMU integration to determine position and attitude of a UXO platform. UWB outages may often occur due to non-line of sight between the UWB network nodes and the rover. To mitigate the errors during short UWB outages, we consider the special dynamics of the platform by adaptively applying constraints in the navigation filter. Typically, a UXO platform moves straight, performs turns or stops; these are the three main dynamic states. Each dynamic state has a set of constraint equations that describes the specific motion. A neural network determines the current dynamic state based on IMU data. Two types of neural networks are examined: (1) a feed-forward network that uses the mean and variance of the IMU data, and (2) a proposed convolution network that takes the raw IMU data as inputs to determine the current dynamic state. The networks are trained on a dataset that was acquired during good GNSS signal reception, and UWB/IMU, GNSS/IMU solutions, whoever, the UWB had some outages. On this dataset, we found that the adaptive constraints mitigate the error of these outage and the UWB/IMU integrated solution by 10% (3-4 cm) using the GNSS/IMU solution as ground truth. The investigation did not show any statistically significant performance difference between the two neural network types.