A loop closure hierarchy to improve the robustness of a wearable vision+inertial navigation system

Critical for the design and development of future space exploration mission vehicles is the effective quantification of astronaut 3D movement and orientation within a long-duration microgravity environment. The International Space Station (ISS) provides a unique opportunity to capture and quantify the architectural layout and 3D space utilization in a microgravity environment from the astronauts living and working there. The design and architecture of future space exploration vehicles requires critical information gathered from the ISS human-system operational environment in order to provide a minimum net habitable volume (NHV) for these systems. This paper extends previous work, which evaluated multiple data selects collected from walking specified routes and performing representative “tasking” in the ISS mockup in the Space Vehicle Mockup Facility (SVMF) at Johnson Space Center. Previous work evaluated Draper's Mischief algorithm, a robust vision+inertial algorithm that includes online sensor-timing error estimation, rigorous parameter optimization, and tuning for many operational environments and use cases to improve the robustness. Draper has developed a more robust state estimation algorithm, Smoothing And Mapping With Inertial State Estimation (SAMWISE), which replicates the vision+inertial navigation capability of the Mischief algorithm, but is capable of additionally incorporating “loop-closure” measurements. In this paper, we describe a two-tier loop-closure hierarchy. The first tier constrains short-duration movements such as gazing back and forth or hover-in-place modes to reduce estimator drift during short-duration but rapid motions. The second tier uses a landmark database to bound the long-duration drift, effectively decoupling the position error of the estimator from usage time when operating within a bounded environment. By leveraging the high-accuracy odometry output of SAMWISE when generating the landmark database, we avoid the need for any time-consuming survey or mapping process prior to operation in a new environment. We present results demonstrating that a landmark database generated by estimating a trajectory from a short traverse through a representative environment can be used to bound the error growth for subsequent and much longer trajectories.