Why Tesla Walked Away from Radar and LiDAR to Go All-In on Vision

In the high-stakes race to solve autonomous driving, there’s a deep philosophical and engineering divide that has emerged over the years.

On one side stands virtually the entire automotive and tech industry, championing a concept called sensor fusion — a belt-and-suspenders approach that combines cameras, radar, and LiDAR to build a redundant, multi-layered view of the world.

On the other side stands Tesla, alone, making the bold and controversial bet on a single modality — pure, camera-based vision.

Tesla’s decision to actively remove and disable hardware like radar from its vehicles was met with widespread skepticism, but it was a move born from a deeply held, first-principles belief about the nature of intelligence, both artificial and natural. To understand just why Tesla took this bet, one must first understand what exactly Tesla rejected.

What is Sensor Fusion?

The concept of sensor fusion is fairly simple. It aims to leverage the unique strengths of different sensor types to create a single, unified, and highly robust model of the environment around a vehicle. Each sensor type has its own advantages and disadvantages, and, in theory, fusing them mitigates each type’s individual weaknesses.

Cameras provide the richest, highest-resolution data, seeing the world in color and texture much as a human does. They can read text on signs, identify the color of a traffic light, and understand complex visual context. Their primary weakness is that they can be degraded by adverse weather and low-light conditions, and they struggle to measure relative velocity.

Radar is excellent at measuring the distance and velocity of objects, even in terrible weather. It can “see” through rain, fog, and snow with ease, but its weakness lies in its lower resolution. No matter how you do the math, it would take a 12-foot by 12-foot square radar array that would cost millions to have the same resolution as a single camera, in a singular direction. It is good at telling you that something is there and how fast it’s moving – as long as it is moving – but it struggles to identify what something is, and struggles to identify objects at a standstill.

LiDAR operates like radar but uses lasers, creating a precise 3D point cloud map of the environment. It is highly accurate in measuring distance and shape, enabling it to construct a highly detailed 3D model of the environment. Its primary weaknesses are its high relative sensor cost and its performance degradation in adverse weather conditions, particularly fog, snow, and rain. LiDAR also has another weakness: the amount of data pulled in is so large that sorting through it requires immense computational effort for the first step alone.

This is the established industry approach, used by companies like Waymo and Cruise, to fuse data from all three, creating a system with built-in redundancy.

Where Tesla Started: A Multi-Sensor Approach

It’s a forgotten piece of history for many, but Tesla did not start with a vision-only approach. Early Autopilot systems, from their launch through 2021, were equipped with both cameras and a forward-facing radar unit, supplied by companies specializing in automotive sensors, like Bosch. This was a conventional sensor-fusion setup, in which the radar served as the primary sensor for measuring the distance and speed of the vehicle ahead, enabling features such as Traffic-Aware Cruise Control and early iterations of FSD Beta.

This multi-sensor approach was the standard for years. Even as Tesla developed its own custom FSD hardware, the assumption was that radar would remain a key component, a safety net for the burgeoning vision system. Then, in 2021, Tesla made a radical pivot.

The Pivot: Why Tesla Abandoned Radar

The shift began in the summer of 2021, when Tesla announced it would remove the radar from new Model 3 and Model Y vehicles and transition to a camera-only system called Tesla Vision. The move was driven by a core, first-principles argument from Elon Musk about the dangers of conflicting sensor data – an argument he continues to make today.

Lidar and radar reduce safety due to sensor contention. If lidars/radars disagree with cameras, which one wins?

This sensor ambiguity causes increased, not decreased, risk. That’s why Waymos can’t drive on highways.

We turned off the radars in Teslas to increase safety.…

— Elon Musk (@elonmusk) August 25, 2025

Elon’s argument is that sensor fusion creates a new, more dangerous problem: Sensor Contention.

When two different sensor systems provide conflicting information, which one does the car trust? Which sensor is considered the “more precise” or “safer” sensor? Is it up to the car in the moment? Is it something decided by the engineers in advance? Sensor ambiguity poses a risk because the decision-making element can be paralyzing, especially when safety is prioritized.

This isn’t just a philosophical argument, either, and Tesla’s FSD engineers have provided concrete examples. In the same thread, Tesla AI Engineer Yun-Ta Tsai noted that radar has fundamental weaknesses – it cannot properly differentiate stationary objects that cannot produce frequency shifts, objects with thin cross-sections, or objects with low radar reflectivity. This is the source of the infamous phantom braking events that plagued Tesla in the past, where a car might see a stationary overpass or discarded aluminum can on the roadside and mistake it for a stopped vehicle, causing it to brake unnecessarily.

From Tesla’s perspective, the road to a generalized solution to vehicle autonomy is to solve vision. Humans drive with two biological cameras and a powerful neural network. The bet here is that if you can make computer vision work perfectly, any other sensor is, at best, a distraction, and at worst, a source of dangerous ambiguity.

Where We Are Today: The Vision on Vision

Today, every new Tesla relies solely on Tesla Vision, powered by its eight cameras. The system uses a sophisticated neural network to create a 3D vector-space representation of the world, which the car then analyzes and navigates within.

The story about vision has a curious footnote. When Tesla launched its Hardware 4 (now AI4), the new Model S and Model X vehicles were equipped with a new, high-definition radar. However, in a move that solidified their commitment to the vision-only path, Tesla has never activated these radars for use in FSD.

In fact, FSD is actually the most evolved on the Model Y, Tesla’s most ubiquitous vehicle, rather than the ones with the additional sensors. While Tesla likely does gather some data from those radars and validates system performance, they aren’t actually part of the FSD suite.

A Binary Outcome

Tesla’s decision to abandon sensor fusion is the biggest single differentiator between its approach to autonomy and that of the rest of the industry. It is a high-stakes, all-or-nothing gamble, which they’re definitely winning so far.

Tesla, Elon, Ashok, and the Tesla AI team all believe that the only path to a scalable, general-purpose autonomous system that can navigate the world with human-like intelligence is to solve the problem of vision completely. If they are right, they will have created a system that is far cheaper and infinitely more scalable than the expensive, sensor-laden vehicles from competitors.

If they are wrong, they may eventually hit a performance ceiling that can only be overcome by those very sensors – but so far, we’ve seen neither hide nor hair of such a ceiling.

Today, Tesla is all-in on its vision-only system, and no one can deny its progress or capability.

For households with multiple Teslas, or for fleet managers, the daily pocket-pat for the right key card is a familiar, if minor, friction point. 

The Model Y card looks identical to the Model 3 card, leading to that awkward B-pillar tap-of-shame when you invariably grab the wrong one. But here’s the thing – you don’t need a separate key card for each vehicle.

A single Tesla key card can be programmed to authenticate multiple vehicles. This simple, elegant capability isn’t a hack or workaround; it’s a fundamental feature of Tesla’s access control setup. Here’s why this works, and how to set it up yourself.

Key Card Versus Phone Key

The first thing to understand is the technological difference between your phone key and your key card.

Your phone key is a smart device. It uses a complex, bi-directional cryptographic handshake over Bluetooth. Your phone and your car are actively paired, creating a secure, authenticated bond specific to your Tesla Account on your phone and the vehicles that have been granted access.

Your key card, on the other hand, is technologically a dumb device. It’s a passive Near Field Communication (NFC) transponder, similar to a modern hotel key or access fob.

The card itself stores no data about your car. It doesn’t know your VIN, your settings, or even that it is a key. All it does is store a single, unique, read-only identifier (UID).

When you tap the card, the car’s NFC reader powers the card for a split second, and the card shouts its unchangeable UID. The car’s computer then checks that UID against its internal access control list of authorized keys.

If the UID is on the list, the car unlocks. If not, it stays firmly shut. That’s why one card can open multiple cars. The card is just the “key.” The vehicle is the one that decides whether to unlock or remain locked.

How to Add Your Key to Multiple Vehicles

The process itself is super simple – and in fact, it’s no different than the process that you would normally use to add a new key card to your vehicle. It just needs to be repeated for each vehicle you’d like the key card to open.

You’ll need your new master key card and at least one, existing, already-authenticated key for each car, whether it be a phone key or another key card.

When you go in your vehicle, navigate to Controls > Locks > Keys, and then tap Add Key. 

Scan your new key card according to your vehicle’s instructions, which differ slightly in where to place the card during authentication.

Once that’s done, repeat the process for each additional vehicle. That single master card in your pocket will now unlock all of your vehicles.

In our ongoing series exploring the technology behind Tesla’s FSD, we’ve covered everything from the universal translator that Tesla uses to map FSD to different hardware, to the clever data pipelines that help automate labeling. We’ve also looked at exactly what the car processes in the environment around it.

What we haven’t talked about yet is just how the car sees the world. Two patent applications from Tesla help to provide this critical detail.

For any real-world autonomous system, two fundamental challenges must be solved. First, how do you accurately measure an object’s distance and velocity? Second, how do you process the massive amount of visual information from multiple high-resolution cameras that see both near and far, without needing an entire supercomputer cluster in every vehicle?

While many competitors try to solve these problems with thousands of dollars in additional hardware and complex sensor-fusion solutions, Tesla has always relied on vision and has a unique way to handle them. Let’s dive in.

Solving for Depth

The first patent, titled “Estimating Object Properties Using Visual image Data”, lays out the key method for why Tesla doesn’t rely on LiDAR except for validation. The core idea is to create a massive training dataset. 

Tesla’s training dataset consists of millions of miles driven by everyday customers, supplemented by the validation engineering vehicles. The validation vehicles use an array of auxiliary sensors to provide highly accurate ground-truth measurements of precise distance and velocity, which are used to help train FSD.

Tesla then uses an automated process to teach the vision neural net. As a validation car drives, it captures a time series of camera images and corresponding auxiliary data. By tracking a vehicle or object across multiple frames, the system can resolve ambiguities, like two cars being close together or one partially blocking another, and correctly associate the precise auxiliary sensor data with the correct car in the image data.

This process generates a massive, highly accurate dataset that Tesla then trains its FSD vision neural network on. It enables FSD to infer depth and velocity from 2D images with a precision close to that of auxiliary sensors. Once the model is trained to a high degree of accuracy and validated, it can be deployed across the entire fleet of customer vehicles, eliminating the need for expensive validation hardware to perceive the world around them.

This is the essence of Tesla Vision: replacing costly physical sensors with a powerful and intelligent neural network.

Solving for Efficiency

The second challenge is managing the immense amount of data from multiple high-resolution cameras without overwhelming the car’s computer. A second patent, “Enhanced Object Detection for Autonomous Vehicles Based on Field of View”, shows that Tesla has worked their engineering magic with another elegant solution.

Processing a full-resolution image from a forward-facing camera is computationally expensive. The common solution is to downsample the image to a lower resolution, but this makes it difficult to detect small, distant objects or read details such as speed signs. A car that is clearly visible at 200 meters might become an unidentifiable smudge of pixels, or a sign that says 80 may be read as 30 in a downsampled image.

Tesla’s method gets the best of both worlds with a technique similar to how the human eye works. The system identifies a priority field of view in the image – typically a horizontal strip near the horizon, which is where distant but important objects are more likely to appear.

FSD then performs two tasks in parallel:

1. It analyses a high-resolution crop of just this priority section, allowing it to see faraway objects with perfect clarity

2. It analyzes a downsampled, lower-resolution version of the rest of the image to efficiently detect closer objects that don’t need the additional resolution

These two processed views are then fused, giving the vehicle a complete picture of its environment that is both long-range and computationally efficient. In computer rendering terms, this is known as foveated rendering – but it is being applied in reverse here. This foveated approach allows FSD to focus computational power where it matters most, which is critical to make a scalable vision system work without lugging around an entire compute cluster on every vehicle.

A Unified, Scalable Solution

Together, these two patents provide a clear view of how Tesla is implementing its Vision-only strategy. Tesla is solving the hardest problems of autonomy not just by adding more hardware to fill the gaps, but by architecting a more intelligent and efficient software stack from the ground up.

If you enjoyed this article, we recommend reading our full series on Tesla patents related to FSD: