What Autonomous Trucks Can’t See
Standalone | Freight Culture Standard | July 2026
An autonomous truck sees the road through three kinds of sensors: LiDAR, radar, and cameras, stacked together because none of them is good enough alone. Each one has a way to fail. Rain and fog degrade some of them by physics. A person with the right equipment can fool others on purpose. Both are documented, tested, and real. Neither is theoretical.
The Physics of Bad Weather
A 2023 peer-reviewed study out of the Korea Institute of Civil Engineering and Building Technology ran controlled road tests measuring exactly how much LiDAR performance drops in rain and fog. The results were not subtle. Reflective materials held up well, retaining better than 74 percent of normal detection even in intense rain and thick fog. Aluminum and steel targets, the kind of material used in ordinary road signs, dropped to zero detection at 20 to 30 meters under those same conditions. The degradation was statistically significant, not a marginal effect.
Radar handles weather better than LiDAR or cameras because of simple physics. Radar’s longer wavelength passes through rain, fog, snow, and dust particles that scatter and absorb LiDAR’s laser light and obscure a camera lens. That is the whole advantage. The cost is resolution. Radar produces a much sparser, lower-detail picture of the world than LiDAR does. It can tell a system something is out there and roughly how fast it is moving, but not the fine shape and edge detail LiDAR provides. That tradeoff is exactly why every serious autonomous trucking system fuses all three sensors together instead of picking one. Weather resilience in radar does not solve the underlying problem. It manages it.
What the Companies Admit
Aurora’s own 2025 Driverless Safety Report, filed through the National Highway Traffic Safety Administration’s (NHTSA) Voluntary Safety Self-Assessment program, states plainly that the company’s Operational Design Domain would expand to include harsh weather. Read that carefully. It means harsh weather sat outside the domain at the moment Aurora launched driverless operations. That is not a hidden admission. It is the company’s own filing.
A named industry representative went further in mid-2024, comparing autonomous truck weather caution to an airline deciding not to fly, and stating that deployment in ice and snow specifically might take a couple more years. Kodiak has tested its trucks in Texas thunderstorms and extreme heat. Gatik partnered with Goodyear on tire technology built specifically to improve safety in adverse weather. Torc Robotics, Daimler’s autonomous trucking unit, said outright that continued testing and sensor improvements are still needed to strengthen performance in inclement weather. Every company that has spoken publicly about this agrees on the same point: weather is not a solved problem. It is a managed limitation, actively being worked on, with more work left to do.
Bot Auto is the exception. No public weather-specific operating limitation was found for the company anywhere in this research.
The Vulnerability Nobody Talks About
Weather is an honest problem. Nobody is trying to cause it. The second category is different, and it gets far less attention in public conversation about autonomous trucking: every major sensor type on these vehicles has been deliberately fooled by researchers using equipment built for the purpose.
Academic researchers spoofed a Velodyne VLP-16, a LiDAR unit widely used across the autonomous vehicle industry, using off-the-shelf hardware. They could inject fake objects into the sensor’s field of view, saturate its ability to detect anything in a given direction, or make an object appear closer than the actual spoofing device. One demonstrated consequence: a spoofed nearby vehicle can trigger a hard brake on a moving truck.
Radar is not immune either. Researchers successfully jammed and spoofed the radar system on a Tesla Model S, creating a false ghost vehicle in the sensor’s field of view and disrupting how the system measured distance to the vehicle ahead of it. Ultrasonic sensors, the kind used for parking and reverse maneuvers, have been spoofed to hide a parked vehicle from detection entirely. Cameras are vulnerable too. A widely cited academic study showed that a physically modified stop sign, altered with stickers a human driver would barely notice, could make a detection model see nothing there at all.
The Distinction That Matters
None of this happened in the field. Every attack described above was a controlled academic demonstration, conducted by researchers who built the equipment specifically to test whether it could be done. No evidence exists of an actual malicious spoofing or jamming attack against a deployed autonomous truck operating in commercial service. That distinction is not a footnote. It is the difference between a demonstrated vulnerability and a documented incident, and the two should never be presented as the same thing.
Manufacturers know these vulnerabilities exist and are building against them. Hesai, a major LiDAR manufacturer, publishes technical documentation on what it calls proactive interference rejection: individually encoding each laser beam so the sensor only accepts returns that match its own signal, filtering out both accidental interference from nearby LiDAR units and deliberate spoofing attempts. Newer LiDAR generations use a related technique called pulse fingerprinting. Academic researchers note it is not foolproof under every condition, but it is a real, engineered response to a real, demonstrated problem.
Who’s Disclosing Anything
Transparency across the four companies most active in autonomous trucking is not even. Aurora has published a specific, named cybersecurity approach: Zero Trust Architecture, meaning every user and system component is continuously verified before it gets access to anything, and cryptographic attestation, meaning hardware and software components are checked for authenticity to prevent tampering. It covers both what is on the truck and what sits off it in the cloud and data systems supporting it. No other company reviewed here has published anything comparable.
Bot Auto’s November 2025 commercial insurance program includes a dedicated cybersecurity policy, which confirms the company treats the risk as real and material, but that is an insurance disclosure, not a technical one. Kodiak and Gatik have not published a dedicated cybersecurity technical document either. All three fall under the same explanation: NHTSA’s Voluntary Safety Self-Assessment program encourages cybersecurity disclosure. It does not require it. What gets published depends entirely on what each company chooses to share.
The Bottom Line
Two different problems, two different natures. Weather is physics, disclosed by most of the companies working on it, actively being engineered around, with real limitations still openly acknowledged. Sensor spoofing is a demonstrated capability, not a deployed threat, built by researchers to prove a point before anyone with worse intentions gets there first. Both deserve attention. Neither deserves exaggeration. The honest version of this story is complicated enough on its own.
Sources
Kim J, Park B, Kim J. Empirical Analysis of Autonomous Vehicle’s LiDAR Detection Performance Degradation for Actual Road Driving in Rain and Fog. Sensors, March 9, 2023. Korea Institute of Civil Engineering and Building Technology.
Deep Camera-Radar Fusion with an Attention Framework for Autonomous Vehicle Vision in Foggy Weather Conditions. PMC, 2023.
LiDAR vs. Radar: Comprehensive Guide and Comparisons. ifm, current as of 2026.
Sensor Fusion for Level 4 Autonomy: LiDAR, Radar, Camera. PatSnap, April 22, 2026.
Prep Underway for Full Driverless Trucks on Texas Highways. Government Technology, March 31, 2025. Source for Aurora’s Operational Design Domain weather language.
Advancing Autonomous Trucking: Kodiak Robotics and Aurora Innovation in Lancaster, Texas. Smart City Consultant, July 4, 2024. Source for the ice and snow deployment timeline quote.
Data vs. Instinct: How Autonomous Trucks Operate in Stormy Conditions. Trucking Dive, June 1, 2021. Source for Kodiak’s thunderstorm testing and the Torc Robotics statement.
Top 5 Autonomous Trucking Companies in the US (2026). Fifth Level Consulting, March 30, 2026. Source for the Gatik and Goodyear tire partnership.
Bishop R. “Disciplined Innovation:” Kodiak Robotics Releases Safety Self-Assessment. Forbes, June 12, 2020. Source for NHTSA’s Voluntary Safety Self-Assessment program background.
Cao Y, et al. Adversarial Sensor Attack on LiDAR-based Perception in Autonomous Driving. arXiv, 2019. Source for the Velodyne VLP-16 spoofing demonstration.
Lidar Sensor for Autonomous Vehicles. Hesai Technology, current as of 2026. Source for proactive interference rejection technology.
Overview of Sensing Attacks on Autonomous Vehicle Perception. arXiv, 2024. Source for the Tesla Model S radar spoofing demonstration.
Towards Robust Sensing for Autonomous Vehicles: An Adversarial Perspective. arXiv, 2020. Source for ultrasonic sensor spoofing mechanics.
Aurora’s Approach to Cybersecurity for Autonomous Trucking. Aurora Innovation, August 19, 2025.
Kodiak AI, PACCAR Weigh In on Liability, Insurance Questions Looming Over Autonomous Trucking. Equipment Finance News, November 19, 2025. Source for Bot Auto’s cybersecurity insurance policy.
Todd Daum is a Safety Compliance Advocate with backgrounds in commercial trucking (CDL-A, HAZMAT, Tanker endorsements), fire service (retired Engineer, Deer Park-Silverton Joint Fire District; Mason Volunteer Fire Department), and fleet safety management. He holds OSHA 30-Hour Construction certification and a Federal Secret Security Clearance, and has built FMCSA/OSHA compliance programs from the ground up for commercial fleet operations.