DynoLoc: Infrastructure-free RF Tracking in Dynamic Indoor Environments

Abstract

Promising solutions exist today that can accurately track mobile entities indoor using visual inertial odometry in favorable visual conditions, or by leveraging fine-grained ranging (RF, ultrasonic, IR, etc.) to reference anchors. However, they are unable to directly cater to "dynamic" indoor environments (e.g. first responder scenarios, multi-player AR/VR gaming in everyday spaces, etc.) that are devoid of such favorable conditions. Indeed, we show that the need for "infrastructure-free", and robustness to "node mobility" and "visual conditions" in such environments, motivates a robust RF-based approach along with the need to address a novel and challenging variant of its infrastructure-free (i.e. peer-to-peer) localization problem that is latency-bounded - accurate tracking of mobile entities imposes a latency budget that not only affects the solution computation but also the collection of peer-to-peer ranges themselves. In this work, we present the design and deployment of DynoLoc that addresses this latency-bounded infrastructure-free RF localization problem. To this end, DynoLoc unravels the fundamental tradeoff between latency and localization accuracy and incorporates design elements that judiciously leverage the available ranging resources to adaptively estimate the joint topology of nodes, coupled with robust algorithm that maximizes the localization accuracy even in the face of practical environmental artifacts (wireless connectivity and multipath, node mobility, etc.). This allows DynoLoc to track (every second) a network of few tens of mobile entities even at speeds of 1-2 m/s with median accuracies under 1-2 m (compared to 5m+ with baselines), without infrastructure support. We demonstrate DynoLoc's potential in a real-world firefighters' drill, as well as two other use cases of (i) multi-player AR/VR gaming, and (ii) active shooter tracking by first responders.