25 August—Right now, there is no way for consumer devices to directly measure speed. But a new Swiss start-up could change that with the first portable sensor that can measure speed instantaneously and exactly.
Lugano-based Vissee has developed a sensor that takes a fundamentally new approach to measuring speed. Vissee’s speed sensor, dubbed "Third Eye," is a neuromorphic processor modeled on the motion detection system of the fruit fly. Only a year into the company’s existence, manufacturers of a wide range of products are eager to get their hands on its prototypes: Next March, Vissee will announce a partnership with a major semiconductor manufacturer; by the end of 2011, the sensor will be in sports equipment; in 2012, the sensor will be integrated into traffic and surveillance micro-aerial vehicles made by Zurich-based Skybotix, according to its CEO, Samir Bouabdallah; and farther in the future, the sensor may be used by a major automotive manufacturer.
To understand why so many disparate companies are courting Vissee, you must first understand how most speed sensors work today.
Vehicle speed sensors, such as those that give system feedback for antilock braking systems, are located between the axle and the wheel. As the wheel rotates, the sensor measures the position of 48 to 60 evenly spaced spokes. This method is accurate to about five kilometers per hour, says James Sterling, an engineer with Infineon Technologies North America who is based in Livonia, Mich.
While a driver can safely use this information to match (or exceed) posted limits, the speedometer is inexact. The error comes from the need to multiply the rotational speed of the spokes by the radius of the wheel, because radius varies with the make and model of the car and the type and wear of the tire treads.
It’s not just cars that measure speed indirectly. A standard smartphone sports app has two ways of measuring speed: either by integrating acceleration or by taking the derivative of position. An app that tracks running times, for example, might calculate average speed with a GPS-provided location. But GPS has a refresh rate of only about once every second—sprinters and skiers would need something more granular—and it doesn’t do well with vertical changes. Alternately, the app could integrate the output of an accelerometer. But this method tends to compound any errors.
An ideal speed sensor would accurately and instantaneously calculate speed independent of position or acceleration. Vissee’s Third Eye will be the first application-specific processor for consumer goods to do so, and on a power budget that CEO Nicola Rohrseitz says is tiny. (To keep competitors guessing, he declines to reveal the power figures.) Rohrseitz and Valeria Mozzetti—Vissee’s founders and both graduates of the Swiss Federal Institute of Technology at Zurich (ETHZ)—created both the hardware and the algorithms by reverse engineering the flight patterns of fruit flies.
As students, they engaged fruit flies in behavioral tasks in which the flies calibrated their speed by observing objects passing by. The objects were really just images projected on the walls of a low-velocity wind tunnel. As the flies hovered in the middle of the tunnel, Rohrseitz and Mozzetti studied their reaction to controlled visual cues. Tens of thousands of such experiments later, they reverse engineered what motion computations the fly’s visual system must have performed. "This was the blueprint for the design of the chip," says Steven Fry, a professor at ETHZ who supervised Rohrseitz’s work in the lab.
Here’s how it works. The sensor is composed of a fish-eye lens, a high-quality 60-hertz CMOS camera, and an ARM-based microprocessor running a special sauce algorithm that selectively filters incoming data. Both the lens and microprocessor are modeled on different aspects of a fruit fly’s ability to navigate.
The lens gives the camera a field of view of 180 degrees, approximating the vision of a fruit fly. (A standard lens can image perhaps 40 or 50 degrees.) The lens funnels light to the CMOS camera. After the visual data is filtered through Vissee’s proprietary algorithm, the image data is substantially reduced, and manageable for the microprocessor.
The chip is looking for two variables: temporal frequency (variation of a signal over time in one point) and spatial frequency (variation of a signal such as light intensity in space in one moment). Approximating a division of the temporal frequency by the spatial frequency can yield a remarkably spot-on measure of absolute speed.
Temporal frequency is easy, but spatial frequency is trickier and more computationally expensive. That means it burns a lot of power—anathema to a small, portable sensor.
Vissee’s algorithm lightens the computational load by using methods derived from the fruit-fly brain to filter out any data that does not help the processors calculate speed.
Rohrseitz envisions his sensor, being used by elite sprinters, skiers, and other athletes to make minor but crucial adjustments to their time. In practice, the Third Eye will be a small, wireless sensor, somewhat like a pedometer.
Farther in the future, Rohrseitz and Mozzetti hope to see their sensor integrated into cars to improve vehicle control. Infineon’s Sterling says automotive suppliers might find the idea interesting for systems such as back-up cameras or adaptive cruise control. But he’s not convinced that Vissee’s sensor would be helpful for currently available antilock braking applications. "Today’s system doesn’t care if it’s doing 20 or 100 when it goes into a skid, either way its brake-pulsing algorithm will be the same." But future ABS system might vary their actions based on speed, and Third Eye could help there.
Massimiliano Versace, a principal investigator on DARPA’s Synapse project—which seeks to draw on neuromorphic principles to build microprocessors that can think and reason in a mammalian fashion—says the Vissee sensor is a good example of how neuromorphic hardware is evolving. Versace says reverse engineering the fruit fly’s visual pathways has solved a problem that has stymied designers of traditional sensors. "This is exactly how to go about using neuroscience to solve otherwise unsolved engineering problems," he says.