- SensorFusion MeetUp (https://www.meetup.com/Sunnyvale-Sensor-Fusion-Meetup/)
- Wednesday, July 25, 2018, presentation 6:30pm to 8pm
- Hosted at Plug-n-play (https://www.plugandplaytechcenter.com/) with over 100 attendees.
- Speaker: Jake Li, Business Development Manager for Hamamatsu Corporation. (jli@hamamatsu.com)
- Slides available (https://drive.google.com/file/d/1RBGGernLF3DwAf9eQKFi9o8rXA7Te9hb/view?usp=sharing)
LIDAR modules for AV are evolving to meet demands by choosing different tradeoffs using mixes of emerging vs. mature technology and detection range vs. safety vs. cost. Throughout the talk, tradeoffs are emphasized as careful choices with different components for different modules, possibly in the same vehicle. No perfect LIDAR module exists at the moment; development is active.
Hamamatsu Photonics (https://www.hamamatsu.com/) provides optical components such as detectors, emitters, and MEMS mirrors to many LIDAR module manufacturers. Hamamatsu also provides components to medical, dental and other industries.
Cameras, radar, and LIDAR accomplish different tasks in autonomous vehicles by measuring different frequencies in the electromagnetic spectrum. Cameras use natural light to provide excellent color and contrast during daylight; radar penetrates adverse weather such as rain or fog; LIDAR provides excellent angular resolution to distinguish objects at a distance. Each sensor overcomes their individual limitations. For example, radar cannot distinguish between two cars within two meters at a range of a hundred meters but LIDAR can.
Water absorption of electromagnetic frequencies in the LIDAR range drives power, safety, and price trade-offs. Lower frequency LIDAR, around 905nm, allows lower cost components (silicon photodetectors and CW/pulsed laser diodes) but are restricted to lower power for eye safety and have consequent lower range. Higher frequency LIDAR, around 1550 nm, provides lower ambient noise; allows lower cost optic components; and can operate safely at high power with increased range. But this comes with poor adverse weather performance, higher-cost InGaAs (Indium-Galium-Arsenide) photodetectors and high power fiber lasers.
A motor on the bottom spinning mirror mounted on the top of a car provides the current mature technology for full circle detection at long range. High cost; parts availability in quantity; large out-of-car mounting; and worries about wear from a moving system are its tradeoffs.
The rotating multi-facet mirror uses a polygonal mirror with each facet reflecting a different angle. It provides a mature technology and long range with tradeoffs of lower field of view and resolution and, again, mechanical wear issues. It can used for typical ADAS systems today.
Flash LIDAR illuminates a larger section of a scene at once, such as an entire scan line. It provides solid state design, compact and lightweight packaging at the cost of being immature for longer ranges. Field of view is traded against the cost of its components. Improvements in VCSEL (https://en.wikipedia.org/wiki/Vertical-cavity_surface-emitting_laser) beam divergence would make this a long range solution.
Beam Steering with MEMS mirrors can be either a 2D scan with a single photodetector and 2D MEMS mirrors; or a 1D scan using a flash scan source (from Flash LIDAR) with 1D MEMS mirrors and an array of detectors. Mirrors are mechanical, requiring checking for resonant frequencies and triggering ISO 26262 concerns. Within 2D mirrors, a choice between electrostatic, piezo, or electromagnetic actuators must be made. In the near future, Hamamatsu foresees a good resolution, range, and resolution solution.
Beam Steering via Optical Phased Arrays provides a true solid state solution by dynamically steering light beams without moving parts. While compact and lightwight, inherent light loss may restrict it to short and mid range. The technology is still in development.
FMCW LIDAR with heterodyne optical mixing borrows concepts perfected in the radar space to provides distance and velocity in a single reading. While lack of ambient noise in this system eliminates the need for a high gain photodetector, it requires more signal processing power and still requires a beam steering technology. The technology is still in development.
The presentation also gave quick overviews of light emitter and photodetector component tradeoffs.