Your behavior appears to be a little unusual. Please verify that you are not a bot.


Following the Team into Danger

June 1, 2013  - By

Ma-opener

An Enhanced Personal Inertial Navigation System

When a team of firefighters, first responders, or soldiers operates inside a building, in urban canyons, underground, in foliage, or under the forest canopy, the GPS-denied environment presents unique navigation challenges. An enhanced personal inertial navigation system (ePINS), based on a strapdown navigation solution using a mid-grade IMU and wavelet-based motion-classification algorithms, can track positions with errors of less than 2 percent of distance traveled in both indoor and outdoor environments.

By Yunqian Ma, Wayne Soehren, Wes Hawkinson, and Justin Syrstad

Numerous pedestrian navigation applications are currently available or proposed for development. Some of them include localization for coordinating firefighters, first responders, or soldiers. In these applications, the safety and efficiency of the entire team relies directly on the location and orientation of each team member. Operations in high signal interference areas such as cities, rugged terrain, forest, or indoor spaces deliver intermittent or no GPS signal. An alternative to GPS-based location is required.

In this article, we introduce an enhanced personal inertial navigation system (ePINS) solution specifically designed for environments where GPS is unavailable. ePINS combines an array of state-of-the-art sensors and fusion algorithms into a personal navigation system that provides accurate location information for pedestrian applications.

The ePINS concept.

The ePINS concept.

The ePINS solution has the following benefits:

  • Accurate positioning in GPS-denied environments;
  • Small, lightweight unit can be easily carried by first responders, rescue workers, or soldiers;
  • Ruggedized packaging to withstand difficult first responder and military environments.

Features of  the ePINS unit include:

  • State-of-the-art micro-electromechanical systems (MEMS) gyros and accelerometers, barometric altitude sensor, and advanced navigation software;
  • Advanced motion classification algorithms that accurately identify and measure user activity;
  • Immunity to magnetic disturbances.

Related Work

In the field of personal navigation, it is common to find systems that rely on sensors that need infrastructure (for example, Wi-Fi positioning) or sensors that actively emit electro-magnetic radiation (such as Doppler radar). These requirements are major drawbacks for communities such as dismounted soldiers in hostile environments.

Other approaches exploit the so-called Zero-velocity update (ZUPT) mechanism, which resets the inertial measurement unit (IMU) velocity errors during the stationary phase of motion. However, implementation of such schemes relies on sensors embedded in footwear, which is not readily accepted in many user communities.

To address these drawbacks, Honeywell has been developing advanced aiding techniques for personal navigation that do not rely on infrastructure and compute a self-contained, relative-navigation solution based only on passive sensors. One technique that Honeywell has developed uses displacement estimation from human-motion models. This technology has been implemented in the ePINS prototype and shows promising performance.

The human-motion model uses IMU measurements as inputs and was developed to infer distance traveled. It generates a displacement estimate that is used as a measurement in the navigation filtering process. The first version of this model was matured under the DARPA individual Precision Inertial Navigation System (iPINS) program. The iPINS system used an IMU, GPS, barometer, and motion classification to estimate a person’s position in both indoor and outdoor environments. In this system, IMU signal characteristics (e.g., peaks and valleys in the accelerations induced by walking) were exploited to differentiate between walking and running. Honeywell recently expanded the human-motion model to identify more specific motion types using a new wavelet motion classification method.

System Description

Figure 1 displays the hardware architecture of the ePINS, a small battery-powered, highly integrated electronic system. The ePINS processing platform is an ARM11-based, i.MX31 system-on-module, paired with support electronics. In addition to the processing platform, the ePINS assembly includes a MEMS IMU, a barometric pressure sensor, a digital magnetometer, and a GPS receiver.

ePINS hardware architecture.

Figure 1. ePINS hardware architecture.

The MEMS IMU provides inertial measurements for strapdown navigation. The IMU’s small package size, light weight, low power consumption, and impressive performance make it attractive for use in the ePINS system. The device is less than 5 cubic inches and weighs less than 0.35 pounds. It consumes about 3 watts of power with a typical current draw of 600mA at 5V.

The ePINS software system is shown in Figure 2. The navigation software runs within Honeywell’s Embedded Computing Toolbox and Operating System (ECTOS IIc), which provides a layered, customizable, and reusable software architecture for implementing navigation, guidance, and control software. A Honeywell-developed simulation tool for offline analysis and development of ECTOS-based software was also used in ePINS development and testing.

Figure 2.  ECTOS IIc hierarchical software structure.

Figure 2. ECTOS IIc hierarchical software structure.

The ePINS demonstration device can achieve path performance of less 2 percent distance traveled for walking motion after 1 hour of operation, independent of the magnetic environment. Current performance, packaging characteristics, and interfaces are summarized in Table 1.

table 1  ePINS performance objectives and physical specifications.

Table 1. ePINS performance objectives and physical specifications.

Algorithm Description

Figure 3 depicts the overall sensor integration and data processing scheme used in the ePINS device.

Figure 3. Sensor integration using the ECTOS extended Kalman filter.

Figure 3. Sensor integration using the ECTOS extended Kalman filter.

Extended Kalman Filter (EKF).  The EKF estimates the navigation and sensor errors and computes the resets applied to the strapdown navigation solution to increase its accuracy. Error models for the navigation sensors (IMU, barometric altimeter, magnetometer, GPS, and motion classification) are contained in the EKF. For the ePINS device, the virtual measurements from the step-length model and the strapdown navigation solution are fused by the EKF to assist in bounding the time dependent error growth of the strapdown navigator, which in turn helps maintain calibration of the inertial sensors. A key output of the EKF is the navigation confidence, which is an estimate of the accuracy of the navigation solution.

An important aspect of the EKF and step-length modeling is the residual test that the EKF supports. This test provides a reasonableness comparison between the step-length model estimate and the distance predicted by the strapdown navigation system. This capability significantly increases the robustness of the navigation solution, especially when the user is engaged in motions not recognized during motion classification.

Human-Motion Model. The human-motion model includes two components: wavelet motion classification and step-length model estimation. The wavelet motion classification identifies the type of motion the user is performing, and the step-length model acts as a virtual sensor that quantifies the motion as a distance-traveled estimate.

Wavelet Motion Classification. Human motions are very diverse and highly irregular. Determining what motion is being performed is a challenging problem of classification. Honeywell’s solution is based on wavelet transformation of IMU data. Predefined, or known, characteristics of a variety of motions (such as walking, running, crawling, etc.) are cataloged and stored to a device’s memory. Estimates of those same characteristics for a user are then computed in real time and compared to the catalog of stored information to find the best match.

Generating the catalog of stored information is an offline task that begins by “segmenting” recorded IMU time domain data into individual steps. An example of the output of the segmentation process is shown in Figure 4.

Figure 4. Segmentation of the IMU data using the y-axis accelerometer signal.

Figure 4. Segmentation of the IMU data using the y-axis accelerometer signal.

Figure 5 displays the segmentation results for two different walking styles (in red and blue) across approximately 15 example steps. As is evident from the graph, walking has characteristics that are common across users, for example, the sharp peaks in the z-axis acceleration caused by foot-ground impacts. Once the data has been segmented, a wavelet transformation on each data channel is performed. Wavelet transformation for many users over many different motion types takes place offline. Subsequently, a wavelet descriptor is built for each motion type based on the transformations into the wavelet domain. With this method, a wide variety of information (that is, descriptors) suitable for input to a classifier is captured about each motion. These descriptors are then cataloged and stored in memory on the ePINS device.

Figure 5. Sample steps for two subjects (red) and (blue).

Figure 5. Sample steps for two subjects (red) and (blue).

Finally, for the online phase, the wavelet descriptor of the incoming IMU data is calculated by performing a wavelet transformation on each data channel. This descriptor is then compared to the pre-computed and stored descriptors to classify the motion. FIGURE 7 shows an example of the motion classifier output, where a running motion was used as an input. The classifier successfully determined the motion type (blue field), frequency and phase of the input motion, depicted by the tallest rectangle in the figure.

Figure 7. Classification results from a query of running at a certain frequency and phase (depicted by the dark sphere).

Figure 7. Classification results from a query of running at a certain frequency and phase (depicted by the dark sphere).

Step-Length Modeling. Once the current motion is identified, a step-length model specific to that motion is used to aid the navigation algorithms. The model for each motion type is obtained by first collecting data that measures step length and step frequency. From this data, the step-length models can be computed by performing a regression analysis of the step-length vs. step-frequency data. Since the step-length models act as a virtual sensor, the models must be as accurate as possible to achieve better system performance. To attain model accuracy, an accurate data collection method is needed.

For ePINS development, step-length models for multiple users have been identified from step-length and timing information using a precise GPS truth reference system. Step-length regression calculations then determine the step length as a function of step frequency (that is, inverse of the step time period).  An example of GPS truth data and the corresponding regression model are shown in FIGURE 6 for walking motions.

Figure 6. Step length versus frequency for the walking of subject.

Figure 6. Step length versus frequency for the walking of subject.

Although basic step-length models are created offline, online calibration of the step-length model can be performed by the EKF if GPS is available during operation. Online calibration tends to increase the overall position accuracy, as variations in the step-length models are likely due to slight variations in biometric differences across humans, terrain features, and even mission plans and duration.

Heading Determination. Heading initialization is one of the key concerns during system start up. In its current operational use, the ePINS device may perform a dynamic or a static initialization of heading. The static method requires the user to survey the system’s initial heading to an accuracy value that is usually specified by mission performance objectives; the absolute position accuracy is dependent upon the accuracy of the initial heading.

The dynamic method is a general method for heading initialization; it is performed without input from the user, but is possible only when GPS is available. This method of heading initialization does not use any a priori information about heading and requires an EKF implementation with a large-azimuth error model. This method requires an additional period of time in which the heading error uncertainty converges.

User Interface. During a mission, the user can interact with the navigation system and monitor its output on a display. The current ePINS prototype offers two-way communication via a serial connection. The serial communication is made wireless by the addition of a Bluetooth interface. Users can use this link to monitor the status of the navigation solution and to send commands to the device.

Honeywell has developed an application for the Android platform for this purpose. One of the key features of the interface design is that the navigation system outputs data in a standard NEMA format. Thus, publically available Android applications, not just proprietary applications, can also receive and display the navigation solution output by the ePINS device.

Honeywell’s personal navigation application displays the user’s traveled trajectory in real-time. The application can be adapted to include building floor plans as well as other navigation information.

Results

The ePINS prototype has been evaluated both in simulations and indoor/outdoor experiments. The navigation results presented here were obtained in February 2012 at a Honeywell facility (FIGURE 8). First, the user completed the heading calibration, and then online step parameter estimation in the presence of GPS was performed. Once calibration and training was completed, the GPS was disabled to simulate a GPS-denied environment outdoors. The user than transitioned to indoors (with GPS still disabled), and walked a course inside that included walking up and down stairs (FIGURE 9) and ended in a conference room (FIGURE 10).

Figure 8. Course for the Honeywell facility demonstration.

Figure 8. Course for the Honeywell facility demonstration.

Figure 9. The user walking up stairs.

Figure 9. The user walking up stairs.

Figure 10. The user at the end of the demo.

Figure 10. The user at the end of the demo.

Over these conditions, the ePINS system performed robustly and within performance specifications. Live demonstrations and testing showing similar levels of performance were performed at the 2012 Joint Navigation Conference (JNC) and at military test sites in California and Indiana.

Summary

The technical approach of the ePINS solution to the problem of personnel navigation in GPS-denied environments is based on a strapdown navigation solution maintained using a mid-grade IMU and advanced motion-classification algorithms. We integrated an array of sensors and software into a system that provides accurate position information and is suitable for use by first responders, soldiers, and other personnel where GPS is unavailable. ePINS works well for a variety of pedestrian motion types, including walking, running, crawling, walking upstairs, walking downstairs, sidestepping, and walking backwards. The motion classification and modeling method is extensible to other motion types.

We tested the ePINS system in indoor and outdoor environments. FIGURE 11 depicts the future ePINS concept, and TABLE 2 presents its future physical characteristics.

Figure 11. Future ePINS concept and mounting position.

Figure 11. Future ePINS concept and mounting position.

Table 2. Packaging characteristics of the future ePINS.

Table 2. Packaging characteristics of the future ePINS.

Acknowledgments

This article is based on a presentation made at ION GNSS 2012.

Manufacturers

The ePINS processing platform uses Honeywell Agile Navigation and Guidance Integrated Electronics support electronics. It includes a Honeywell HG1930 MEMS IMU, a Bosch Sensortec BMP085 barometric pressure sensor, a Honeywell HMC6343 digital magnetometer, and a NovAtel OEMStar GPS receiver.


Yunqian Ma is a principal scientist at Honeywell Aerospace. He received his Ph.D. degree in electrical engineering from the University of Minnesota, Twin Cities. He is currently the program manager of the GPS-denied navigation program and the next-generation personal navigation program.

Wayne Soehren is a senior technical manager at Honeywell Aerospace. He was the program manager for the development of Honeywell’s first MEMS-based GPS/INS, which developed the core capability now used in Honeywell’s IGS-2XX family of MEMS-based GPS/INS products. He holds an MSEE from the University of Minnesota.

Wes Hawkinson is an engineering fellow at Honeywell Aerospace. He holds a BSEE/CE from the University of Wisconsin–Madison.
Justin Syrstad is a guidance and navigation scientist. He received a master’s degree in aerospace engineering from the University of Minnesota.