Consumer Robotics: A Platform For Embedding Computer Vision In Everyday Life

Consumer robotic devices provide a platform for embedded computer vision algorithms in applications for everyday life. The consumer market is very price-sensitive, so robots must be developed with a single task in mind, aiming to provide the best performance at the lowest cost. Computational resources in consumer robotics are scarce given cost constraints, forcing the design of novel algorithms that elegantly incorporate such constraints. We present a graph-based SLAM approach designed to operate on computationally constrained platforms using monocular vision and odometry. When computation and memory are limited, visual tracking becomes difficult or impossible, and costs for map representation and updating must remain low. Our system constructs a map of structured views using only weak temporal assumptions and performs recognition and relative pose estimation over the set of views. We fuse visual observations and differential measurements in an incrementally optimized graph representation. Using variable elimination and constraint pruning, graph complexity and storage is kept linear in explored space rather than growing over time. We evaluate performance on sequences with ground truth and also compare to a standard graph-SLAM approach.