MapLearn: Indoor Mapping using Audio
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Indoor mapping, signal processing, machine learning
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Cameras and LIDARs are established methods to generate the map (or floorplan) of an indoor environment. This paper investigates the feasibility of using audio to learn the map. We aim to transmit audio beacons from a mobile device (say a smartphone) and record its reflections from the environment. Assuming known user locations, and recordings from multiple locations along walked paths, we aim to learn the 2D floorplan of the area. We use a conditional GAN (cGAN) architecture but prevent it from over-fitting using knowledge of indoor signal propagation. We pre-train our model on simulated data -- thousands of high-fidelity audio measurements on hundreds of synthetic floor plans -- and then test on 4 real environments in our home and office buildings. Results show that the generated maps are fairly accurate (in terms of precision and recall) even though no training was performed in real rooms. We have assumed clutter-free rooms; coping with clutter remains a topic for continued research.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5801
Loading