The emerging era of highly-automated driving comes courtesy of much-vaunted sensor technology—spinning lasers, penetrating radar and sonar blips. But the sensor that gets the least respect in the technology stack could be the lynchpin for the mass roll-out of self-driving vehicles: the inertial measurement unit, or (IMU).
“IMUs are the glue that binds everything together,” said Mike Horton, chief technology officer at Aceinna, a Boston-based company that develops sensing solutions for automotive applications. "An IMU can help bring different types of data together from GNSS (Global Navigation Satellite System), lidar and cameras to make everything consistent and smooth.”
Moreover, an IMU can mean the difference between life and death. The unit's main components—an accelerometer and gyroscope—determine a vehicle's velocity, acceleration, heading and turn rate. IMUs provide a reliable “ground truth” even when there’s a gap in GPS, a vehicle heads into a blind corner, or lidar struggles with a snowstorm.
“When other sensors fail, the IMU is the sensor that gets you to the side of the road to a safe spot,” said Horton. “It's uniquely positioned to do that because it has no external dependency.”
New tricks from an old tool
In its basic form, an IMU uses three linear acceleration sensors at X, Y and Z axes, as well as rotation sensors at three axes. By adding a magnetometer, an IMU can include references to earth's magnetic fields to expand from six degrees of freedom to nine. However, given common magnetic disturbances, the magnetometer primarily is used to refine gyroscope readings.
The use of physical gyroscopes goes back more than a century. But wide use in the modern era came about with the development of micro-electro-mechanical systems (MEMS). Today’s miniature accelerometers and gyros are soldered to circuit boards and strapped into cars in enclosures about the size of a matchbox. Basic IMUs are used to detect sudden deceleration to deploy airbags and enable stability control (they also orient your smartphone).
IMU sensors for airbag deployments are available for tens of dollars from traditional auto suppliers. At the other end of the spectrum, sophisticated IMUs from Honeywell and others are used in high-precision aerospace and tactical military applications. The step up from MEMS to fiber-optic gyroscopes (FOGs) brings greater accuracy, but FOGs cost tens of thousands of dollars.
Critically, IMUs also run software—with some providers primarily focusing on algorithms for sensor fusion. Aceinna’s open-platform IMU, for one, allows end users to run their own algorithms.
Unfortunately, the vague nomenclature applied to IMU variants creates confusion. The lines between IMUs and inertial navigation systems (INSs) easily get blurred, making it difficult to make apples-to-apples comparisons of specifications and cost.
“IMU is an umbrella term to describe a wide assortment of inertial systems,” said Jaya Krushna Panda, an industry analyst with Technavio, a London-based market-research firm. Technavio pegs the global automotive IMU sensor market at U.S. $1.13 billion in 2018—mostly for rudimentary safety functions rather than automated- driving; the number of leading IMU providers targeting vehicle-automation applications is about a dozen.
Automated driving changes everything
“Going forward, our sensors will continue to get smaller and more accurate,” said Erica Zelazny, sales director at Xsens, a provider of inertial technologies for sensor fusion based in the Netherlands, with its Americas headquarters in Los Angeles. "Of course, everybody also wants them to be cheaper," she added.
Horton, Aceinna’s CTO, agrees that requirements for accuracy, size and price are going into overdrive. “I've been surprised how aerospace-like the requirements have been for IMUs in autonomous driving,” said Horton. “The big U.S. automotive companies want really high performance.” At the same time, he said the heart of the IMU automotive market remains at less than $500 per unit.
One of the biggest challenges is to correct for gyro drift, often indicated as “in-run bias stability” measured in degrees per second. “No matter what, a gyroscope will drift over time in terms of a zero point," said Scott Kimbrell, senior field applications engineer at Xsens. Engineers also battle the effects of sensor “noise,” such as angular random walk and bias estimate. These factors could result in a vehicle moving toward an adjacent lane (or worse) when there are gaps in input from GNSS or other sensors.
Horton said the goal is to maintain 30-cm (11.8-in.) accuracy for between a few seconds and a couple of minutes. “One degree per hour is the sweet spot that we see in autonomous now," he said. Aceinna’s OpenIMU300 series, which typically offers six-degree-per-hour accuracy, sells for $200 or less.
Xsens’ newest line of IMUs includes the MTi-600, a high-end unit with advanced features and a gyro capable of 8 degrees per hour; it sells for $300 when purchased in quantities of at least 500 units. The company says the in-run bias stability for its gyros will continue to improve, but an accurate MEMS gyro is less important than the power of sensor-fusion algorithms to generate highly accurate readings for pitch, roll and yaw.
Meanwhile, MEMs manufacturing continues to improve. While a one-degree-per-hour IMU is not yet commercially available at automotive-grade prices, Aceinna is literally inching closer to the goal, promising a late-June introduction for a new IMU offering about two degrees per hour for less than $100.
Continue reading »