Categories
Autonomous Vehicles, New Mobility & the Built Environment

The Integrated Driving Sensor – Far, Fast, and Precise

The following YouTube videos and associated descriptions are from a recent interview with Behrooz Rezvani, Founder and CEO of Neural Propulsion Systems.

It is clear that NPS is looking beyond the era of human driving, where up to 94% of crashes are caused by human misbehavior/error.  In the driverless era, insurance will shift from insuring drivers to insuring products.

Rezvani points out that society is less forgiving of mistakes by machines as compared to human errors.  To build trust in robotic driving systems, robotic chauffeurs are going to have a safety record similar to that of the aviation industry.

This zero-accident future is going to require reliability and redundancy.  It is also going to require cost-effectiveness so that it can be deployed at scale. With its approach of integrating solid-state components and advanced signal processing, NPS appears to be well along the path that leads to safer driving.

As an aside, it was a pleasure to catch up with Behrooz after so many years and to see our paths intersect once again around lasers (and more).* In many ways, today’s autonomous vehicle world is similar to our time together in the early days of last-mile broadband and video on demand.

Article outline

The Premise – A Zero Accident Future #

“You need to extend the range, but on the short-range…. you need to pick up subtle movements. It is similar to sports,” states Behrooz Rezvani. With this analogy, Rezvani, Founder and CEO, of Neural Propulsion Systems (NPS), describes a fundamental requirement for getting to, in his words, a zero-accident future.

This requirement to see long and fast at the same time is inherent in NPS’s newly introduced integrated sensor suite and data fusion machine, the NPS 500. OEMs focusing on driverless (i.e., SAE Level 4/5) are the target market for this first-generation roof-top-mounted hardware device.

The Road to Zero #

Rezvani explains that the way to improve an autonomous vehicle’s perception engine is by providing it reliable data and time to make decisions. Overlapping, solid-state sensors increase reliability and reduce calibration requirements, compared to earlier-generation mechanical components. Giving the perception engine more time means extending the range and increasing the frame rate of the sensors.

This translates into the following high-level requirements

  • detecting objects greater than 500 meters
  • sensing whether vehicles are potentially not going to comply with traffic signs
  • seeing pedestrians, particularly children, around corners and behind vegetation

To this last point, AAA has repeatedly warned of the limitations of current pedestrian detection methods, particularly in their ineffectiveness at night and their inability to see around corners.

A Time Machine of Sorts #

Rezvani describes what he and his team are building as a time machine with a resolution measured in picoseconds. Using relatively low-cost solid-state lasers similar to what might be found in data centers, NPS applies multi-input, multi-out (MIMO) techniques to focus the lasers and detectors to illuminate the field of view where the action is (e.g. movement). It is the mathematics, as Rezvani terms it, on the processing side that is able to lift the signal from the noise.

Similarly, their SWAM™ (Sparse, Wideband, wide-Aperture, Multi-band) radar also uses MIMO to steer the beams. With four bands, ranging from UHF to 77 GHz, they are able to see long distances and through vegetation, while being able to discern high-resolution at closer distances. Rezvani likens it to having a telescope and a microscope. Again, it is the signal processing that does the heavy lift, integrating the images from the four different bands, pulling the real signals from ghost images, and tracking objects.

In addition to the 360-degree lidar and radar coverage, the two-piece NPS 500 includes nine cameras leaving no blind spots. Although not discussed during the above interview, 5G connectivity is inherent in the NPS-500. The multi-sensor approach NPS is taking lends itself to further integration, which leads to lower costs and increased reliability.

Better Sensor Performance #

According to NPS, some of the things that make their approach unique include:

  • 360-degree sensing with no blind spots.
  • Multiple sensors (lidar, radar, cameras) provide redundancy.
  • Multiband radar, in combination with its software, allows the NPS 500 to “see around the corner” in the city and up to 1,000 meters on a country road. This approach provides “70X better resiliency against other radar signal interferences.”
  • The MIMO (Multi-Input, Multi-Output) LiDAR provides a range of 500 meters while reducing the potential for interference from other light sources.
  • Up to 2- and 3.3-times faster Radar and LiDAR frame rates compared to industry-leading systems, respectively, allow for improved motion detail compared to industry-standard components.
  • A variable frame rate that goes as high as 100 frames per second is more power-efficient while providing greater performance than systems with fixed frame rates of, say, 20 frames per second.
  • 100% solid-state means no moving parts improving reliability & enabling self-calibration.
  • A combination of custom signal processing integrated circuits and software that filters the 650 Tb/s peak sensor data to provide a fused data stream of relevant information, discarding unnecessary data. The upshot is that this reduces the demands on the OEM’s perception engine, potentially allowing for lower-cost and lower power, third-party processors.

Although raw sensor data is available from the NPS 500, the fused data potentially lightens the load for the perception engine’s processing requirements. They also save power by varying the frame rates based on the action at a given time.

Overcoming Interference Challenges #

Radar is the solution for seeing through objects. But, what happens when every vehicle has radar and is beaming signals into the ether of a foggy road? Similar to how modern WiFi routers are able to direct signals to specific devices, Rezvani explains that their combination of 360° aperture on the roof, MIMO, and multi-band will provide superior rejection of radar signals from other vehicles.

In a subsequent email exchange, Rezvani explains,

“based on the extensive modeling we have done, and significant amount of data published in cellular network design, our two lower radar bands will get thru concrete buildings with windows.”

He describes how all four bands of radar coordinate with each other while scanning deep into a cross-section.

“As the vehicle moves towards the cross-section the position of the angle of the car relative to the cross-section changes. This change of the Field of View (FoV) and continuous monitoring of the Region of Interest (ROI) sampled repeatedly over time (for example every 1 msec) provides a wealth of information. By the time we get to the cross-section, there are well over 1000 radar snapshots. These snapshots provide statistically relevant information for our A.I. algorithm to eliminate ghost targets and stay focused on the real targets.”

Self Calibration and Low Maintenance #

The system’s design is to make it easy for NPS’s customers, the OEMs, to manufacture and to ensure low-cost operation. Connection to the vehicle only requires a 12 Volts DC and a connection to the vehicle’s perception engine,

Rezvani points out that the failure of one component does not render it inoperable. Further, it is modular to the extent that it is possible to replace failed components. Finally, self-calibration reduces operating costs.

Locating the sensors on top of the roof, instead of burying in windshields and bumpers, has the added advantage of potentially eliminating the high-costs associated with repairs. For instance, Jacques Amselem, Head of IoT for Allianz Technology, mentioned at the March 4th, 2021 Smart Driving Car Summit a two-year study that found “that the replacement cost of a windscreen (windshield) with integrated stereoscopic cameras is ten times higher than a traditional glass windscreen.”

The Road to Commercialization #

OEMs, the companies that build and sell companies are the primary target.  He projects samples to these customers by the end of 2021 or 2022. In the 2023 or 2024 timeframe, Rezvani expects the product to be hardened and production-ready. Long-term, NPS’s goal is to make it affordable enough for all vehicles. He sees it as potentially even serving the after-market. He sees it eventually leading to sensing systems that add $4,000-5,000 cost to a vehicle.

Rezvani has been down this road before and understands the difference between prototypes and mass-production, as early in his career he dealt directly with automobile OEMs. As a founder of successful start-ups in the communications world (Quantenna, Ikanos) and with an academic background in lasers, Rezvani brings the technical and operational credibility to create a commercially viable product. But more than commercial success, it is clear that Behrooz’s bigger mission is to make mobility safer for all.

*In the 1990s, this author and Behrooz worked at Raynet, which was a pioneer in early FITL (Fiber In The Loop) systems and video on demand efforts.

Author Ken Pyle, Managing Editor

By Ken Pyle, Managing Editor

Ken Pyle is co-founder of Viodi, LLC and Managing Editor of the Viodi View, a publication focused on independent telcos’ efforts to offer video to their customers. He has edited and produced numerous multimedia projects for NTCA, US Telecom and Viodi. Pyle is the producer of Viodi’s Local Content Workshop, the Video Production Crash Course at NAB, as well as ViodiTV. He has been intimately involved in Viodi’s consulting projects and has created processes for clients to use for their PPV and VOD operations, as well authored reports on the independent telco market.

Linked In Profile

One reply on “The Integrated Driving Sensor – Far, Fast, and Precise”

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: