June 7, 2018
Aside from Tesla, almost every other major company developing automated driving systems has concluded that lidar sensors are going to be a necessary to creating a robust sensing suite. However, this laser-based technology has been prohibitively expensive until now. Making these sensors commercially viable requires driving down the cost and making them durable enough to survive life on the road. One of the leaders in developing the solid-state sensors needed to achieve these goals is Israeli startup Innoviz which has just announced BMW as its first production customer.
Founded in 2016, Innoviz has been developing a micro-electromechanical system-based (MEMS) solid-state lidar system. The company is currently producing its InnovizPro sensor that is designed as a stand-alone unit that can be added to existing vehicles. It is primarily being utilized for automated driving development applications but Innoviz plans to continue offering this product.
The BMW deal is for the next-generation InnovizOne sensor which is specifically designed to be automotive grade and integrated into new vehicle builds. The One sensor takes advantage of what has been learned already from users of the Pro sensor to both improve performance and dramatically reduce cost. For example the field of view goes from 73×20 degrees on the Pro to 120×25 degrees on the new sensor. The frame rate and range also improve from 20 fps and 150 meters to 25 fps and 250 meters.
Itamar Roth, VP of business development for Innoviz declined to be too specific on the cost of his company’s sensors. However, he did acknowledge that the InnovizPro currently costs in the “low thousands of dollars” and the InnovizOne “will be well under $1,000 for its product cycle” with the exact cost depending on production volume. That’s still not inexpensive, but far less than the $4,000 Velodyne is charging for its latest VLP-16 sensor. General Motors is using five of those Velodyne sensors on each of its automated Chevrolet Bolts.
In addition to the hardware, Innoviz is providing its own computer vision software platform that will be integrated into the core stack being developed by BMW. The Innoviz software is designed to analyze the lidar point cloud and recognize targets of interest, classify what they are and track them. The vision software is claimed to be able distinguish vehicles, cyclists, pedestrians and even road lane markings when the reflectivity is sufficient. This will provide an important complement to the optical camera sensors that are currently used for lane detection.
The recognition software also contributes to the dynamic scanning of the laser. Innoviz can change the scanning pattern of its MEMS mirror system for varying conditions such as when the vehicle is operating in a dense urban environment with a lot of nearby objects or when it is driving at higher speeds on the highway to provide information about what is down the road. When objects are detected, the laser can target them to provide more detail for the perception system.
Like most lidar makers, Innoviz is utilizing 905 nanometer lasers which are available at a much lower cost than the 1550 nm lasers used by Luminar. The longer wavelength lasers are considered safer for eyes, but they are also much more expensive. Because they are eye-safe, 1550 nm lidar sensors can operate at higher power levels for increased range. According to Roth, Innoviz has a solution that makes its 905 nm lasers safe while retaining the range performance but he declined to provide details.
The actual processing of the sensor data will take place in the central compute platform along with all the perception, path planning and control operations. BMW is partnering with Intel and its subsidiary Mobileye to develop its automated driving systems. Suppliers Magna International and Aptiv are also working with BMW. Magna is responsible for the overall hardware integration into the vehicle while Aptiv is handling the electronics integration.
Production of the InnovizOne sensors begins in 2019 with manufacturing partner Jabil Optics although Roth says the company plans to work with additional suppliers as well. Production is expected to quickly ramp up to thousands per month. While BMW is the first OEM announced for production applications, Innoviz is working with other automakers and expects to make additional announcements soon.
One of those is likely to be tier one supplier Aptiv. Aptiv is developing its own full-stack automated driving systems and is also an investor in Innoviz along with two other lidar startups, Quanergy Systems and Leddertech. In addition to offering its turnkey driving package to OEMs, Aptiv is also developing its own automated mobility service offerings and partnering with Lyft. The longer range InnovizOne is likely to be used as the forward looking sensor with the shorter range units from its other partners around the perimeter of the vehicle.
BMW has not responded to inquiries on this announcement. The automaker has previously indicated that it plans to launch both level 3 conditionally automated and level 4 highly automated systems into production by 2021. L3 systems are capable of providing partially automated driving under certain conditions. The driver has to take over control when the system cannot handle a scenario, but they don’t have to keep hands on the wheel and eyes on the road at all times as they do with a level 2 system. An L4 system is similar to those being developed by Waymo, GM, Ford and others which doesn’t require human hand-off, but still has a limited operating domain.
A number of new lower-cost solid-state lidar sensors are expected to reach volume production in the next two to three years including a sensor debuted by current market leader Velodyne at CES last January. Despite Elon Musk’s claims that lidar is a crutch for automated driving systems, the added coverage they provide when used with cameras and radar will likely make them essential to building public trust in automated driving technology.