Real-Time Dynamic Pose Estimation Systems in Space: Lessons Learned for System Design and Performance Evaluation

Real-time pose estimation systems are fundamentally difficult to evaluate. Pose estimation performance is a function of the object geometry, environmental conditions, line of sight, sensor characteristics, algorithm characteristics, and computational platform. This paper presents lessons learned from Neptec’s two operational space systems for real-time dynamic pose estimation: the Space Vision System used by NASA for assembling the International Space Station, and the TriDAR Autonomous Rendezvous & Docking System used for spacecraft docking to the International Space Station and automated satellite servicing. Lessons learned are discussed in the context of design, trade-offs, challenges, and performance evaluation. This paper also discusses applying the lessons towards adapting these technologies for other applications. These lessons result in a proposed generalized framework for performance evaluation of pose estimation systems. Such a task is not without challenges given the reliance of pose estimation on so many factors external to the system. Shape-based, targetless systems are particularly difficult to evaluate given the potential range of object shapes and views. However, predictable performance is critical for the success of space operations and NASA’s strict requirements have provided many valuable lessons towards performance predictability. Collaborative research is presented on metrics for quantifying shape with respect to pose estimation, providing a potential means for evaluating and predicting system performance with respect to shape properties.