Echo SLAM – How to fly an omnidirectional bat?
In this project, we explore new approaches to problems at the intersection of acoustics and indoor localization. More concretely, we work on theory and algorithms for simultaneous localization and mapping (SLAM) with range-only measurements based on sound or radio waves.
Inspired by the capability of bats to navigate by listening to echoes of their chirps, we propose to use a device equipped with a single collocated source and receiver. The basic idea is that the device traverses an a priori unknown trajectory, and uses echoes to measure its distance from nearby reflectors. The task is then to jointly determine both the trajectory of the device and the geometry of the reflectors. Since the most ubiquitous example of such a device is a smartphone, we expect our work to unlock tremendous opportunities in indoor localization, a topic of great interest to many technology companies. This interest can be explained by the numerous potential applications and the fact that no current technology has been able to address indoor localization well enough to become standard.
To solve the indoor localization problem, we propose the following approach that leverages the often unexploited potential of echoes. Suppose that a mobile device carrying an omnidirectional source and an omnidirectional receiver traverses an unknown trajectory. At a certain location along its trajectory, the source produces a pulse, and the receiver registers the echoes. Sound propagation is described by a family of room impulse responses (RIRs), where each RIR is idealized as a train of Dirac delta impulses corresponding to the echoes, and recorded by the microphone. From each RIR, we can extract first-order echoes and obtain their propagation times. The case of collocated microphone and source is illustrated below.
A particularity of this setup is that the propagation times directly reveal the distances between the measurement locations and the walls. Our recent research shows that the room geometry and the measurement locations can be jointly recovered from only a few noiseless distance measurements; however, this reconstruction is not always unique. Furthermore, we established the conditions under which uniqueness is guaranteed and tested our algorithm in real experiments.