High fidelity modelling and simulation of inertial sensors commonly used by autonomous mobile robots ∗

Abstract. Autonomous mobile robot behaviors are notoriously difficult to predict using simulations. Robot behaviors are driven by the robot’s sensor systems, and sensor outputs need to be produced in simulations at near-truth levels. However, in order to achieve real-time or faster performance, most simulation environments use empircal, effects-based sensor models, and these simplified models do not provide sensor outputs at the level of fidelity necessary to develop and test autonomy algorithms. Simulated sensor outputs must accurately reflect not only the sensor’s intrinsic noise and signal processing effects but also the sensor’s interaction with its external environment. To address this need for physics-based sensor models, models of the Global Positioning System (GPS), micro electro-mechanical (MEMS) accelerometer, and MEMS gyroscope sensors were developed using a first-principles, physics-based methodology. These sensor models run in near real-time on high performance computering (HPC) assets. The sensors were simulated for a test site located in Vicksburg, MS, and the simulated sensor outputs were compared to truth data collected using a Garmin hand held GPS unit and low cost 6 DOF commercial inertial measurement unit (IMU) consisting of three orthogonal MEMS accelerometers and 3 orthogonal MEMS gyroscopes. The sensor models were also used as part of a simulated autonomous unmanned ground vehicle reconnaissance mission, and results from this simulation are presented