Matt LibertyAutonomous systems that navigate their surroundings are becoming more viable as sensor technology improves and computing power increases. Google’s car is the most recent high-profile system, but even vacuum cleaners and lawn mowers now drive themselves. These systems track their linear and angular position to move through the environment. Some solutionsoperate in an unknown environment and must also perform obstacle detection and avoidance. A variety of sensors exist for autonomous system designers including linear accelerometers, angular velocity sensors (gyroscopes), magnetometers, pressure sensors (barometers),global positioning system (GPS), cameras, IR rangefinders, laser scanners, speedometers and beacons. No single sensor produces the full linear position and angular position, and the current best-practice technique combines the outputs from multiple sensors using a process called “sensor fusion”.

Sensor fusion is a statistical mathematical method for computing the best estimate in the presence of noisy or inaccurate measurements. Calculating the best estimate is an optimization problem, and the Kalman filter, developed by Rudolf Kalman in the 1960s, is the traditional workhorse. The Kalman filter is the optimal solution for linear systems with normally (Gaussian) distributed noise. Estimating angular position is a non-linear problem addressed by the Extended Kalman filter and other variants. Figure 1 shows a typical land-based motion measurement system. 

Figure 1 – Typical Motion Measurement System

Autonomous systems would ideally be self-contained and independent of environmental factors. An inertial sensing system of linear accelerometers, gyroscopes and magnetometers isa common solution. Hillcrest’s low-cost, low-power solution that combines these sensors with proprietary sensor fusion algorithms in a single compact device is shown in Figure 2.

Figure 2 - Hillcrest's 9-axis Inertial SensorWhile inertial sensors can determine angular position accurately, determining linear position is problematic. Computing linear position requires linear acceleration to be doubly integrated, a numerically unstable process. Integration of noise results in bias, and the integration of bias results in a large linear position error. To further complicate matters, linear accelerometers also measure gravity, and any error in angular position couples into the accurate removal of gravity and the linear position error.

Even though a stable, low-cost self-contained inertial navigation solution is not currently practical, the system can add additional sensors to provide an application-specific solution. Any sensor that measures linear velocity can help provide long-term stability to the linear position outputs. Viable sensors include speedometers and a downward facing camera. A speedometer estimates linear velocity by measuring wheel angular rate multiplied by the wheel circumference. A downward facing camera can compute velocity using frame deltas, similar to an optical mouse. By combining the linear velocity estimate into the sensor fusion process, the system can produce a more accurate linear position estimate.

Adding external reference signals further helps estimate linear position. GPS is the most common solution, but many other reference systems exist. A beaconing solution can use many different technologies including RF, IR and magnetic. The beacon could even be the RF signal strength for existing 802.11 or mobile phone networks.

As sensor technology continues to improve and drop in price, autonomous systems will become more common, but much of the technology is already available. Designers can purchase Hillcrest’s Freespace® Sensor Modules today for systems that can benefit from high performance, low-cost inertial sensing.

About the Author:
Matt Liberty has been a core developer of Hillcrest Labs’ Freespace® motion control technology for 7 years.He has been granted 9 US patents related to motion control and has firsthand experience with motion sensing technologies. Products using Freespace technology include the Logitech MX Air and LG SmartTVs. Matt has a broad engineering background encompassing digital signal processing, PCB development, FPGA development and software development in C, C++ and Java. Before Hillcrest, Matt developed telecom equipment for two other startups: Megisto and Salix Technologies. Matt has anundergraduate degree in electrical engineering from Cornell University.