A huge new West Hall was added to the Las Vegas Convention Center for CES 2022 to accommodate all the automotive technology, and it seemed like nearly every booth had an Autonomy and/or electrification story to tell. Machine vision and sensing are the keys to unlocking the potential of autonomy and eventually ridding our roads of the 90-plus percent of vehicle collisions attributable to human error and to keep our goods moving with truck drivers in short supply. Here were some cutting-edge technologies we spotted—using human vision—on the machine-vision and sensing fronts.
Light Distance And Ranging (LiDAR) Sensors
SiLC Frequency-Modulated Continuous Wave LiDAR
Some 39 machine-vision companies presented LiDAR solutions at CES 2022, but of these SiLC’s Frequency-Modulated Continuous Wave solution impressed us the most. Here’s how FMCW LiDAR works: Instead of determining distance by measuring the time it takes for laser light pulses to go out to an object and come back again (time-of-flight), these lasers continuously vary the frequency (color) of their light. They then split this oscillating color laser beam in two, routing one beam directly to the light collector while the other is projected out into the environment. But before that beam exits, it travels through a spiral or racetrack-shaped “wave guide” that measures about 2 meters in length, but at nano scale it all fits on a fingernail-sized silicon chip. Those 2 extra meters put that beam’s color phasing out of sync by a set amount, against which the total amount it’s out of sync can be used to compute the distance the reflected light traveled out to the object and back.
- Advantages: Continuous illumination greatly increases range (500 meters, vs 200-300 for most others) and reduces power draw (200 milliwatts versus 120-1,000 watts for others), directly measures velocity as well as range, produces a clearer image, covers a reasonable field of view (four headlamp-mounted units can cover the forward field of view).
- Disadvantages: Indium-phosphate lasers cost more than VCSEL-type lasers, but practical automotive pricing is anticipated with scale, technology is lower on the development curve (probably five years out).
Luminar Iris & Hydra
Luminar has the most mature technology, entering its third generation of production. It uses the pulsed time-of-flight method of measuring range, by sending pulses of 1,550-nanometer light to scan the environment, with the laser beam aimed by micro-electromechanical systems (MEMS) mirrors (meaning it’s not technically solid-state). It can detect items at 500 meters, but the more meaningful range number is how far it can identify items that have just 10 percent reflectivity (like a tire in the road), and that number is a still-impressive 250 meters.
Luminar’s low-profile Iris sensor appeared at CES 2022 atop the Volvo Concept Recharge, and it’s expected to power ADAS systems from Volvo, Polestar, and Geely among others. Luminar’s Hydra product was displayed on a semi-tractor. Luminar has announced a LiDAR-only ADAS kit that has yet to be announced for production (Volvo will team Luminar’s LiDAR with cameras and radar.
- Advantages: 1,550-nm light is safer for eyes, which allows the power level (and hence range) to be increased, beam-splitting and micro-mirror scanning provides up to 120-degree field of view.
- Disadvantages: The indium-gallium-arsenide semiconductor material used in the laser is costlier to produce in a hermetic environment (though very little of it is needed) and water diffuses 1,550-nm light more than it does 905-nm light, diminishing effectiveness in inclement weather.
Opsys seems to be leading the machine-vision race in then the solid-state LiDAR class, with a micro-flash approach that mounts two separate arrays comprising hundreds of tiny vertical-cavity surface-emitting (VCSEL) lasers flanking a single collector. By flashing each of these tiny lasers the system scans the scene without moving parts. And by using two different light frequencies on the two chips (905 and 940 nm) Opsys eliminates the possibility of crosstalk, which helps broaden its field of view. It also allows the system to scan the environment very quickly (1,000 frames per second, when ADAS only requires 30 fps) at very high resolution with low enough power to render the shorter-wavelength laser light eye safe at sufficient power levels to reach 300 meters, though the distance at which it’s 90 percent likely to identify an object with 10 percent reflectivity is 145 meters.
The superfast frame rate means that where an object is detected, the chip can order greater resolution in that area while still scanning the rest of the scene at the requisite 30 fps. A level-four autonomous trucking application is on track for 2024 production, and Opsys LiDAR will be incorporated in headlamps produced by Korean supplier SL for unspecified automakers at a likely cost in the $200/unit range at scale.
- Advantages: Gallium-arsenide VCSEL lasers are extremely affordable as are the silicon-based detector arrays, shorter wavelength laser light is absorbed less by water, so it performs better in bad weather.
- Disadvantages: Not many.
Typical radar units that typically power today’s adaptive cruise control systems use a flat radar antenna, but Lunewave proposes distributing and receiving its waves via a Luneburg lens, like the ones the military attaches to stealth fighter jets during training to render them temporarily visible on friendly radar screens. Think of these as fisheye lenses for radar waves. The fighter jet versions might be basketball sized and cost $10,000-$20,000 each, but for automotive duty, Lunewave has shrunk the concept to ping-pong-ball size and proposes producing its intricate 6500-facet shape of acrylic in a 3D printer.
Planar lenses suffer “tunnel vision,” providing clarity over the 10 degrees directly in front, which degrades farther out in any direction. The Luneburg lens helps channel the radio waves in such a way as to produce a clearer image across a broader field of view (up to five times better than traditional radar) at distances of up to 350 meters. This broader, clearer view means two to four Lunewave units might be able to replace the 16-20 planar antenna units required to provide the 360-degree coverage required for full autonomy. The hardware to transmit and receive the radar waves basically is the same, so added cost is minimal. Volume production is anticipated for 2023. Advantages: Improved clarity across wide field of view means fewer units are needed. Disadvantages: The unit cannot mount perfectly flush with the bodywork, though when integrated, it’s not as bad as gluing a ping-pong ball to the front of a modern radar unit.
Adasky Viper Thermal Camera
Promising vision in any weather, the Adasky camera presented at CES 2022 passively records the heat signature from objects in the environment, as opposed to some solutions that illuminate the area ahead with infrared light and look for reflection. This solution “sees” much better at night than visible-light cameras that are dependent on headlamp illumination. Anything living obviously stands out fairly brightly, but there’s generally enough temperature difference to distinguish painted lane divider lines as well, meaning the view ahead typically just looks like a black-and-white, sometimes negative image of what your eyes see.
Adasky can detect and identify pedestrians and animals at 250 meters—day or night—as opposed to 65-120 meters with headlights (low to high beam)—and then only if they’re standing in the light. As forward-collision systems become mandatory, the government is likely to impose performance standards, which no current visible-light camera system can pass at night or when staring directly into sunlight, so Adasky believes demand will soon skyrocket. Cost is estimated at $100 per unit at scale, and there is a contract for 2024 production with a North American OEM.
- Advantages: “Sees” through fog, uses less than 1 watt of power (compared with 5-25W for LiDAR).
- Disadvantages: Cannot produce both visible and IR image, so probably still need both visible and thermal cameras.
Gentex In-Camera Mirror and Structured Light Cabin Monitoring
This new cabin monitoring system include a typical camera that mounts inside the rear-view mirror that is also intelligent enough to discern eye status, hand-held cell-phone use, eating, smoking, etc., all of which help assess and score a driver’s distraction level.
The camera can be augmented with a colinear-light scanning laser chip mounted in the overhead console. Kind of like the Opsys LiDAR unit above, it projects dots of light throughout the cabin, measuring reflected returning light. These reflected light dots can detect micro vibrations that indicate signs of life. It can even detect heart or breathing rates penetrating a rear-facing car seat with a sleeping baby in it. This information is used to warn a driver—either with tones or messages sent to a phone—that a person or pet has been left in the car. This product made its debut at CES 2022, so there are no production plans to report yet.
There’s no machine vision through dirt and grime, so this clever air-based gizmo cleans a sensor using little jets of air generated locally the way a fireplace bellows works. A round, flat chamber, sized to suit the area that requires cleaning, inhales and exhales quickly in response to piezo-electric pucks (the same tech that powers many direct fuel injectors). The design of the opening tends to suck air in from the sides of the opening and blow it out through the center. The effect of this is to speed up the outflowing air and boost its effective pressure.
Other compressed air cleaning systems require a remote compressor and pressure lines that are costly to run and vulnerable. ActJet only needs 12 volts and a signal line to order a cleaning. Speaking of which, cleaning can be scheduled at periodic intervals, perhaps informed by other onboard systems monitoring conditions like dust and rain, or by the sensor itself when it detects an obstruction. ActaJet can be paired with a fluid washing system, serving to accelerate the spray speed to cleanse tougher obstructions like dirt and insects. This startup is engaging automakers and Tier-1 suppliers with no active contracts to report.
- Advantages: Compact, easy to adapt mounting directly to or very near a sensor, lower power draw than distributed compressed air.
- Disadvantages: The team still has some noise/vibration issues to resolve en route to production.
TTP Ultrasonic Sensor Cleaning
This concept is so new we don’t have a photo of it, but TTP, a UK-based independent technology company, has come up with the idea of placing a layer of glass or clear polymer in front of the sensor and vibrating it at ultrasonic speeds that either break the weak bonds that hold dust and dirt to glass, or causing hard particles to bounce, rather than stick, or by atomizing liquids and ejecting them. Similar ideas are in production for removing dust from internal components of high-end cameras. It’s believed that the power required will be akin to that which drives a doorbell or parking sensor.
Source: Read Full Article