Previous Abstract Return to Session D1 Next Abstract

Session D1: Robotic and Indoor Navigation

Anatomical Joint Constraint based INS/INS fusion for Indoor Position Estimation and Tracking
Gerald LaMountain, Mohsen Nabian, M.Veysi Yildiz, Amirreza Farnoosh, Pau Closas, Sarah Ostadabbas, Deniz Erdogmus, Taskin Padir, Northeastern University
Location: Spyglass

The development of strapdown inertial navigation techniques and algorithms has vastly improved the number of applications in which inertial navigation technologies may be utilized by allowing inertial navigation systems to function in spite of many of the physical constraints of the systems into which they are included. In particular by using more robust and inexpensive inertial measurement unit (IMU) devices instead of gyrostabilized accelerometer systems, strapdown inertial navigation systems eliminate the problem of gimbal lock, effectively increasing the rotational freedom of the sensors that comprise the system, and by extension that of the components of the physical systems to which they may be attached. One particularly interesting example of such an application is in robotic and human indoor navigation, in which sensors must be attached to limbs and joints which may have a high freedom of motion resulting in the ability of the robotic or human body to execute complicated maneuvers and assume a wide variety of poses. In addition to the challenges inherent with working with a physical system which exhibits such rich kinematic behavior, the indoor aspect of these applications can result in intermittency or connection unreliability with global navigation satellite systems (GNSS), which are traditionally fused with dead reckoning inertial systems to correct for inertial bias and system unobservability in position estimation applications. This means that there can be relatively large periods of time in which position estimation solutions will be subject to divergence from the true position of the sensor.
One possible approach to address these challenges is to make use of the fact that the dimensions and limitations of the physical system can be measured and known a priori. Limbs are of typically of fixed, measurable lengths, and joints are limited in the direction and distance in which they can rotate. We hypothesize that by applying constraints based on these known physical quantities and kinematic limitations of robot or human bodies to an algorithm based on traditional strapdown equations and techniques, the navigation solution are improved in terms of system accuracy and robustness to dropouts in GNSS position estimation. Several hybridization schemes can be possible, for which one can think of approaches conceptually similar to those from classic IMU fusion like loose or tight coupling. As a first approach to the general goal of optimally fusing body-mounted IMUs, this paper will develop the theory considering a pair of sensors that have generic physical constraints which relate their relative positions, velocities, and attitudes. We will discuss, through computer simulations, a variety of two-joint configurations featuring different constraints on the rotational freedom of each of the joints. This is a crucial step towards designing more general multi-IMU fusion algorithms, for which we will study two-joint configurations that are relevant also in the context of body constraints. Following this, we will test the application of these constraints to a particular physical toy case in which a double pendulum with IMU sensors attached to the body of each segment of the pendulum is suspended from a robotic (aerial) platform which is allowed to move freely in two or three dimensions. The objective is to show how optimal data fusion can be achieved to estimate the location of the platform by using measurements from the two IMUs. We will compare the performance of the constrained solution for each of the two sensors under GNSS and GNSS dropout conditions to those computed without application of those constraints.



Previous Abstract Return to Session D1 Next Abstract