MarvinEye - Autonomy in Constrained Environment

摘要

We present MarvinEye, a novel computer vision-based framework with many possible applications, successfully validated on a semi-guided autonomous rover traversal. The framework is based on an innovative, robust, and extendable three-tier architecture composed of a primary-secondary setup and an elector for outlier and error filtration. MarvinEye is demonstrably accurate to compete with an RP-LiDAR and requires only two algorithms with a dominant elector on a divergence, making MarvinEye a resource optimal alternative to proximity sensors. We present a thorough experimental evaluation demonstrating extremely positive results leading to successful validation of the proposed architecture in the context of a semi-guided autonomous rover traversal among many other potential applications.