We all hope that the next generation will suffer from far fewer tragic accidents, or even better, that they will be a thing of the past. This goal guides us as we move through the second half of 2024, and continue as an industry in our steady march in developing safe and efficient self-driving technologies. New regulatory standards and safety guidelines kicked off the race for autonomous vehicle sensor supremacy into even higher gear, and government authorities are pushing automakers to maximize their sensor arrays to gather as much high-resolution data as possible to train their powerful AI perception systems.
Demand is increasing for higher-performance sensors and data interfaces to enable these powerful systems, which are required for safe driverless operation. These dynamics drive LiDAR companies to work harder in order to meet the demand, yet every automaker sees what’s about to come differently. Here are some of the key trends we’ve seen emerging in 2024 and how they are going to shape our future when it comes to mobility, autonomy, and most importantly – road safety:
Insatiable Demand for More Pixels
One of the biggest areas of focus is increasing the resolution of the sensors that allow self-driving cars to “see” their surroundings. Higher pixel counts provide more detail, allowing smaller obstacles to be detected at greater distances. This enables smoother, faster autonomous driving by giving the AI systems more time to identify and respond to hazards, improving safety as a result.
Camera resolution is pushing into the gigapixel range, with sensors capable of up to hundreds of millions of frames. We see progress in LiDAR too, with point clouds getting denser and top firing millions of times per second.
But more pixels are just part of the equation. Frame rate – the number of images/point clouds captured per second – is another critical factor. Higher frame rates reduce latency, meaning, the delay between when the sensor data is acquired and the vehicle’s reaction to it. With autonomous driving, even milliseconds of latency can matter when it comes to identifying and responding to fast-moving objects or sudden changes.
Many self-driving systems are now using cameras and LiDAR units with at least 10fps, while the best products on the market offer 20 frames per second or higher. Lower latency translates directly into faster reaction times. This also provides more data points for the perception algorithms to work with, reducing false alarms from brief distortions or occlusions in the sensor feed.
Keeping Pace with Soaring Data Rates
With 2024 well into its second half, even Gigabit Ethernet is becoming a bottleneck for the latest high-definition LiDAR and multi-camera rigs. Top automakers are now transitioning to multi-Gigabit interfaces like 2.5Gb Ethernet or GMSL (Gigabit Multimedia Serial Link) and more.
These ultra-fast pipelines are essential to bringing all the rich sensor data to the computing clusters that fuse it into a coherent 3D model of the vehicle’s surroundings. Without sufficient bandwidth, the full potential of the high-resolution scanners can’t be realized. There is simply too much data to stream in real-time over slower legacy interfaces.
New Sensor Types on the Horizon
While LiDAR and cameras continue to push the limits of resolution and data rates, other novel sensor types are also emerging, which have the potential to place additional bandwidth demands on autonomous systems.
Automotive radar is getting a revamp too. New digital radar chipsets offer higher resolutions and faster scanning rates. These evolving radar technologies also offer additional data streams to enhance overall perception systems, with top digital radar units now capable of data rates approaching 1 Gbps.
Securing the Data Torrent
From higher megapixel cameras and denser LiDAR clusters to new sensors and 4D digital radar, it’s clear the autonomous vehicle industry is being inundated with a torrent of data. And with more data comes increased security risks that must be addressed.
As these high-bandwidth sensor feeds are transmitted over vehicle networks, automakers will need robust cybersecurity measures to prevent snooping or tampering. Encrypting the raw data flows and authenticating every component will be critical to safeguard this sensitive information that could potentially be exploited to remotely interfere with a self-driving system.
Data storage and processing is another security minefield. The massive datasets used to train autonomous driving AI models contain an incredibly detailed record of vehicles’ travels that could pose major privacy risks if mishandled. Strict data governance policies, secure cloud pipelines, and access controls will all be essential to protect this proprietary and personal information.
Moreover, the centralized compute clusters making real-time decisions based on sensor outputs represent an attractive target for bad actors. Rigorous cybersecurity protocols like kernel-level firewalls, containerization, and secure updates will be paramount to prevent these critical systems from being hacked or disrupted.
While the focus remains on improving autonomous capabilities, ensuring the integrity and security of the data driving these AI systems cannot be an afterthought. As data volumes and bandwidth escalate, investing in robust security will be just as vital as the sensors and processors themselves.
To conclude: The Future of Autonomy Driven by Data
From insatiable demands for more pixels and higher frame rates to rapidly increasing bandwidth requirements and emerging security challenges, one thing is clear – the future of autonomous vehicles will be driven by data. Massive volumes of highly detailed sensor inputs are the fuel powering the artificial intelligence systems that will finally realize the long-awaited vision of self-driving cars.
As the autonomous revolution continues, we can expect the data torrent from these sensor arrays to become even more of a deluge. Automakers will need to keep pushing the limits on resolution, frame rates, and data interface bandwidth to gather the incredibly rich training datasets required.
Navigating the autonomous future won’t be easy. But with continued innovation on the sensor and data fronts, combined with rigorous practices around data governance and security, the road ahead leads to fully self-driving vehicles operating with unparalleled levels of safety and efficiency. The autonomous age is fast approaching, powered by the tremendous streams of data that will ultimately teach vehicles to transition from “drivers” to “AI riders.”
Innoviz is developing cutting-edge LiDAR and Perception Software to meet the increasing demands for higher-resolution and faster data processing in autonomous systems. Read more about our InnovizTwo™ LiDAR and Perception Software or contact our Sales team for more information.