NavShoe Pedestrian Inertial Navigation Technology Brief Eric Foxlin Aug. 8, 2006 WPI Workshop on Precision Indoor Personnel Location and Tracking for Emergency Responders
The Problem GPS doesn t work indoors Cannot assume a prepared environment or use of any tracking infrastructure To obtain meter-level accuracy and thorough coverage with RF/UWB requires setting up multiple antennas around the building Pedometer/compass dead-reckoning modules not sufficiently robust and accurate Must calibrate for individual user s step size Changes in size of steps cause errors Variations in direction of steps cause errors Current products specify drift accumulation of 2-5% of distance travelled, and that assumes normal walking with consistent gait.
Inertial Navigation gimbaled stable platform accelerometers gyros motor motor motor
Inertial Navigation - strapdown accels gyros rate gyros ω B integration orientation accels f B coord. transform f N a N double integration -g position
Inertial Navigation error growth 500 450 commercial-grade tacticalgrade strategicgrade 400 350 navigation-grade geophysical limit mm 300 250 200 150 100 50 0 10 0 10 1 seconds 10 2 10 3
NavShoe Concept Foot-mounted sensor package contains MEMs gyros, accelerometers and magnetometers Short-term inertial navigation measures the 6-DOF trajectory of each step works with any kind of motion Break cubic error growth by resetting velocity to zero after each step: Take advantage of correlated position/velocity errors in Kalman filter to also remove most position error with each ZVU: Correct heading drift of small MEMS gyros, based on compass measurements averaged over a long distance
NavShoe Hardware PDA or wearable computer User interface & information display app GPS RF NavShoe data fusion software Wireless InertiaCube3 foot-mounted sensor cluster
Results (Indoors) User walked through a typical wood-frame house for 322 seconds, covering a total distance of 118.5 meters. Started at position (0,0,0) on the first floor in the living room. The final reported x,y,z position of the tracker in meters is (-0.32 0.10-0.06), indicating that over the whole journey it has drifted by 0.3% of the distance traveled. The NavShoe keeps track of the height The total drift in altitude over the experiment was only 6 cm, or 0.06% of distance traveled
Results (Indoors) Trajectory of NavShoe during 118.5 m exploratory path through house. 6 bed upstairs bedroom 4 5 northing (meters) 4 3 2 1 1 st -floor bedroom dining table up (meters) 3 2 1 0 sofa kitchen 0-1 -2-1 -4-3 -2-1 0 1 2 3 4 5 6 easting (meters) -2-1 0 1 2 3 4 easting (meters) Plan view Elevation
Results (Outdoors) Trajectory of NavShoe during 741m road loop. northing (meters) 20 0-20 -40-60 -80-100 -120-140 -160-50 0 50 100 150 easting (meters)
Algorithm for Integration with GPS or LPS Initially, inertial heading and magnetic declination both set with high covariances As the user walks, heading and declination become highly correlated. When GPS becomes available, we use Transfer Alignment measurements after each step to align inertial heading very precisely to true geodetic North. Because of the high correlation, this allows the filter to make a precise estimate of magnetic declination. During GPS outages, the compass is compensated with declination, and used to keep the inertial heading aligned to geodetic North. Long-term navigation fusing GPS with NavShoe could produce results more accurate than GPS alone
NavShoe Results w/ GPS Outage (Outdoors) User walked a 1059 m closed loop through a very hilly residential neighborhood logging data from the NavShoe and GPS. During the first 400m, the GPS fixes were incorporated into the NavShoe Kalman filter For the remaining 659 meters, the GPS fixes were completely ignored. Data post-processed in Matlab.
NavShoe w/ simulated Denied GPS (Outdoors) 200 NavShoe unused GPS fixes GPS fixes used for training declination northing (meters) 150 100 50 0-150 -100-50 0 50 100 150 easting (meters)
NavShoe vs. Torso Dead Reckoning Requires user calibration/training Requires special algorithms to identify backwards steps Can be fooled by nonstandard motions Torso Dead Reckoning Devices Yes Yes Yes NavShoe No No No Depends on specific orientation on user s body Yes No Typical accuracy 2-5% of distance walked 0.3% of distance walked
NavShoe Issues & Limitations Will users accept shoe-mounted sensor? Maybe as part of a self-powered insole Will the performance hold up for crawling, running and any other maneuvers that firefighters may make? Will performance hold up in real-world temperature extremes? Will drift be low enough to use it without external position fixes?
Future Work Integrate with real-time processor Support crawling and running Variant with sensors on both feet, and possibly inter-foot ranging sensor as well Integrate optical flow, Doppler, or Baro-altimeter sensors Tight integration with GPS Tight integration with SLAM
No Silver Bullet This problem is hard enough to require sensor fusion. InterSense has 10 years experience in Kalman filter integration of inertial and aiding sensors (magnetic, acoustic ranging, optical, GPS ) We have a generalized sensor fusion core architecture with API for easily integrating new sensors. Looking for partners with complementary technologies. Thank You!