TECHNOLOGY AREA(S): Information Systems
OBJECTIVE: Develop a capability to maintain the accuracy, integrity and reliability of tracking dismounted soldier trainees as they perform their exercises in the field.
DESCRIPTION: The Army’s current modernization of live training requires accurate dismounted soldier trainee position/tracking to enable the convergence of Live, Virtual, and Constructive (LVC) training environments. Currently, live soldier trainees, when represented virtually, are often seen to have position jitter, or can be seen floating or jumping to positions not physically possible in the real world. Currently, under conditions of GPS satellite signal attenuation or blockage due to terrain features is conducted by a concept called dead reckoning, which is similar to a flywheel effect, which are relative estimates of position and heading using inertial sensors: accelerometers to measure linear motion, and gyroscopes to measure angular rate change. These sensors estimate position, velocity, and heading measured from a last known trusted GPS receiver position measurement (latitude and longitude) when the satellite signal reception is attenuated, distorted, or blocked by land features such as trees or buildings. Accelerometers simultaneously detect walking steps and estimate stride length to derive an estimate for position and velocity. Gyroscopes measure angular rate changes and are used to estimate heading. The last known GPS receiver position measurement also receives timestamps from received satellite signals that are referenced and tracked by crystal oscillators for keeping time reference measurements to estimate velocity. All of these sensors have error modes that degrade Position, Velocity, and Timing (PVT) measurement estimates proportional to distance travelled and elapsed time. Current state-of-the-art dead reckoning error rate is approximately 2% of distance travelled on flat, even, terrain.We are seeking innovative dead reckoning techniques based on time series-based algorithmic solutions that exploit artificial neural networks that have PVT estimate error equal to, or less than, 0.2% of distance travelled and is able to maintain estimate performance in challenging terrain to include stairs, tunnels, and steep mountain terrain. Solutions using low cost Micro-electro Mechanical Systems (MEMS) based Inertial Measurement Units (IMUs) are preferred to keep cost, weight, and power consumption low.Soldier traineesoften perform their exercises in the field which are often in environments where GPS satellite signals are substantially degraded or altogether unavailable, such as during maneuvers in indoor urban training center buildings.
PHASE I: Develop detailed analysis of predicted performance and perform modeling and simulation of technical approach. Phase I deliverables will include a design concept and analysis of expected performance capability with supporting rationale.
PHASE II: Develop, demonstrate, and validate a proposed dead reckoning system using a Linux Operating System (OS) that meets the topic objectives. Phase II deliverables will be dead reckoning system prototype that can demonstrate meeting topic objectives in an outdoor test environment. The use of an Android smartphone to demonstrate the technology capability is acceptable. The proposed solution must be mounted on the body’s upper torso.
PHASE III: Potential military applications would include dismounted soldier navigation under tactical operational conditions where GPS satellite signals are attenuated or obscured or under electronic warfare situations. Commercial applications would include vehicle fleet tracking, Unmanned Aerial Vehicles (UASs) performing aerial surveying data collections in GPS-challenged environments to maintain public safety, or for wearable gait analysis to detect changes in the neural control of gait linked to ageing or Parkinson’s disease.
KEYWORDS: Machine Learning, Neural Networks, Time Series Forecasting, Kalman Filter, and Pedestrian Dead Reckoning
1. Q. Wang, L. Ye, H. Luo, “Pedestrian Stride-Length Estimation Based on LSTM and Denoising Autoencoders”,Sensors 2019, https://www.mdpi.com/1424-8220/19/4/840;2. Q. Wang, L. Ye, H. Luo, “Pedestrian Walking Distance Estimation Based on Smartphone Mode Recognition”, Remote Sensing 2019, https://www.mdpi.com/2072-4292/11/9/1140;3. G. Guo, R. Chen, “A Pose Awareness Solution for Estimating Pedestrian Walking Speed”, Remote Sensing 2019, https://www.mdpi.com/2072-4292/11/1/55;4. C. Chen, C. Lu, B. Wang, “DynaNet: Neural Kalman Dynamical Model for Motion Estimation and Prediction”, https://arxiv.org/pdf/1908.03918.pdf;5. S. Rangapuram, M. Seeger, “Deep State Space Models for Time Series Forecasting”, NeurIPS 2018https://pdfs.semanticscholar.org/7b8f/97663a190ddaf599f83d3ab6204639f8881a.pdf?_ga2.57350316.789179070.1569364791-670032685.1569364791