Authors: Roshan Ayyalasomayajula, Aditya Arun, Chenfeng Wu, Sanatan Sharma, Abhishek Sethi, Deepak Vasisht, Dinesh Bharadia
Location services, fundamentally, rely on two components: a mapping system and a positioning system. The mapping system provides the physical map of the space, and the positioning system identifies the position within the map. Outdoor location services have thrived over the last couple of decades because of well-established platforms for both these components (e.g. Google Maps for mapping, and GPS for positioning). In contrast, indoor location services haven’t caught up because of the lack of reliable mapping and positioning frameworks, as GPS is known not to work indoors. Wi-Fi positioning lacks maps and is also prone to environmental errors.
DLoc is a Deep Learning based wireless
localization algorithm that can overcome traditional limitations of
RF-based localization approaches (like multipath, occlusions, etc.).
DLoc uses data from the mapping platform we developed, MapFind,
that can construct location-tagged maps of the environment. Together, they allow off-the-shelf Wi-Fi devices like smartphones to
access a map of the environment and to estimate their position with
respect to that map. During our evaluation, MapFind has collected
location estimates of over 150 thousand points under 10 different
scenarios across two different spaces covering 2000 sq. Ft. DLoc
outperforms state-of-the-art methods in Wi-Fi-based localization
by 80% (median & 90th percentile) across the 2000 sq. ft. spanning
two different spaces.
MapFind (left) is an autonomous platform that maps an indoor environment while collecting wireless channel data. The platform generates a detailed map of the environment and collects training data for DLoc. DLoc uses the training data to learn a model to accurately localize users in the generated map.
Once DLoc has been trained using the data collected from MapFind. Any new user entering the environment can use their smartphone to estimate their location with the help of the trained network.
After the map is generated, MapFind goes around the mapped space in an optimal path pre-determined by our path-planning algorithm so as to cover sufficient space in the provided map as shown above.
Simultaneously the server on board the bot collects the CSI information and converts them into the Input heatmaps as shown above. These Input images are then used to train the DLoc network shown above using both LLocation and LConsistency losses. These losses back propogate as shown by the red dotted lines to train DLoc.
While being the first in in Deep Learning based Indoor Navigation with WiFi data, we want to build WiFi CSI dataset on par with ImageNet to assist further research in WiFi based indoor localization and their applications.
The 1st international workshop for Wireless AI perception is happening at CVPR 2022. More details can be found at [WAIP].
April 2022A 2nd iteration of [WILD] has been released for a Kaggle competition. More details can be found at [WAIP].
September 2021PyTorch implementation for DLoc network architecture has been released and can be acccessed at [Codes]. The features we generate from the raw CSI nformation to train the network can be found at [Datasets].
September 2021Sound based localization based on DLoc's implenetation called "Sound source localization based on multi-task learning and image translation network" paper got accpeted to JASA 2021, Vol 150, Issue 5
February 2021Sound based localization based on DLoc's implenetation called "Blind Sound Source Localization based on Deep Learning" paper got accpeted to ICASSP 2021
September 2020We presented our paper at Mobicom 2020. A recording can be found here
August 2020DLoc datasets with the largest open-sourced CSI raw data has been released and can be accessed at [Datasets]
Feb 2020Our paper titled "Deep Learning based Wireless Localization for Indoor Navigation" has been accpeted to appear in Mobicom 2020