Mobile Robotics Lab Advanced Personal Dead-reckoning (PDR) System for Firefighters Presented by Johann Borenstein Research Professor at the University of Michigan 29 years experience in GPS-free position estimation for mobile robots and pedestrians
If you like the GLANSER form factor you re gonna love ours
PDR Packaging IMU is fully embedded in heel of firefighter boot. IMU is easily transferable among boots of different firefighters. Computer & battery kept in pouch on shaft of firefighter boot no need to run IMU-to-computer cable up the firefighter s leg. Battery life: >3.5 hours. Built-in connectivity: Bluetooth WiFi USB, RS-232 900 MHz consumer-grade radio 2.64 (67 mm) Multi-function button Battery test Battery button test LEDs IMU & charger port Radio LED Antenna 4.13 (105 mm)
PDR Features Eliminates effect of accelerometer drift with every step (ZUPT). Uses our patent-pending Heuristic Drift Elimination (HDE) algorithm to eliminate effects of gyro drift and other gyro errors in buildings. Does not require any user-specific calibration whatsoever. Does not require setup of any infrastructure whatsoever on location. Initialization at start of mission takes less than 8 seconds. Demo/test at undisclosed military test site in June 2012. Walk included walking backward, sideways, zig-zag, up and down stairs, intentional walking in arbitrary directions. Return Position Error: ~3 m.
PDR Performance Average heading errors: <1 o in missions of unlimited duration! Average position errors: <1% of distance traveled in walks >30 min. Average elevation (Z-axis) errors: <1.5 m in walks of unlimited duration. Works with running Errors ~2 x walking errors Works with crawling on hands and knees, even up and down stairs Errors ~2-4 x walking errors Works with users changing step-length or gait. Works with users walking/crawling backward or sideways. Battery life: >3.5 hours. End CCAT demo in January 2012 Included one lap each of: Walking, running, crawling, even hopping on one leg Start
What s New since 2011 WPI PPL Workshop Support for running Improved accuracy during crawling New Flexgain feature Initialization time is now < 8 seconds (was 25 s) New Adaptive Footfall Detection Much greater accuracy Much greater robustness Latest News: Customer in Singapore purchased two PDR systems for testing with firefighters. PDR systems were delivered to customer on July 24, 2012, after three days of acceptance testing by customer s representative at our lab.
Live Demo A picture is worth a thousand words (unnamed wise man) A video is worth a thousand pictures (Steven Spielberg) A live demo is worth a thousand videos (Johann) Initialization: Stand still for 7+ seconds Walk 8 steps in parallel to one of the walls of the building Trajectory of live demo walk/crawl performed during this presentation. Lap 1: walking Lap 2: crawling Return position error: ~1.2 m Start Stop
Summary U of Michigan personal dead-reckoning (PDR) system Two key features: 1. Removes drift from IMU s accelerometers 2. Removes drift from IMU s gyros 3. Work will every conceivable kind of legged motion, including crawling Results: Average position errors: <1% of distance traveled in walks >30 minutes Average heading errors: <1 o in walks of unlimited duration (at steady state, indoors, in >99.5% of buildings) Two modes of elevation estimation (average Z-axis error: < 1.5 m) The PDR system is ready for field testing and for integration with other first responder systems. We are seeking industry licensees for our technology
Backup Slides The following slides are not part of the oral presentation but may be useful in the Q&A session
Live Demo of Walk and Crawl at 2011 PPL Workshop Presenter performed a live demo of a short walk and crawl Walk Podium End of walk, start of crawl Crawl Trajectory of live walk (~50 m long). RPE < 0.4 m (<1% of distance traveled) Trajectory of live crawl in continuation of the walk. Total travel, ~85 m long. RPE <3 m (3.5% of distance traveled)
Most Recent Results (July 24, 2012) Running
Most Recent Results (July 24, 2012) Crawling on hands and knees
Most Recent Results (July 24, 2012) Crawling up staircase
Dominant Directions in HDE Algorithm The HDE algorithm is based on the concept of Dominant Directions(DDs) Two sub-sets: primary and secondary DDs Primary DDs are those that one notices immediately when looking at the floor plans of most buildings, and they are usually the same as the directions of the outside walls of the building. Most corridors in buildings are parallel to primary DDs (see floor plan to the right). In rare cases corridors intersect with primary DDs at angles of 45 o. Those are called secondary DDs. Typically, a walk with the PDR system must be confined to just one building. That is because the DDs must be the same throughout a walk. If adjacent buildings have the same DDs, then a single walk may stretch across all of these buildings. Floor plan of the conference center of an unnamed hotel. The general direction of the corridors is highlighted by thick red lines. The four primary dominant directions (0 o, 90 o, 180 o, and 270 o ) are typically those that are parallel to the walls and main corridors of the building. Secondary dominant directions (45 o, 135 o, 215 o, and 315 o ) intersect primary ones at angles of 45 o. 180 o 135 o 90 o 45 o 215 o 270 o 315 o 0 o
What s New Since 2010 PPL Workshop, cont d Floorplan Mode Allows incident commander to define regions that are not rectilinear