A new fleet of Tesla engineering validation vehicles has been spotted in Los Gatos, California, sporting peculiar new hardware. Unlike the standard validation mules we’ve seen for years — usually equipped with roof-mounted LiDAR rigs for ground-truthing — this Cybertruck and Model 3 feature a new, customized camera rig mounted directly to the corners of the front bumper.
The rigs, which appear to be on temporary brackets, position wide-angle cameras low to the ground, angled outward. While Tesla frequently tests new sensors, the specific placement of these cameras - targeting the frontal blind spots that are difficult to capture from the existing camera architecture - suggests that Tesla is collecting data to solve for a pair of their biggest current challenges: Summon and Banish.
Ground Truth
Tesla’s current vision philosophy relies on occupancy networks - software that predicts the presence of objects in blind spots based on what the car saw seconds ago. This memory is crucial for Tesla’s vehicles - but the Cybertruck especially - given the fact that the front-facing cameras cannot see certain areas around the bumpers.
However, memory isn’t enough for precise low-speed maneuvers. These new rigs appear to be establishing a new ground truth dataset. By pairing these low-mounted corner cameras with high-fidelity roof LiDAR, Tesla is likely capturing a pixel-perfect map of the ground immediately around the bumper.
This data will be used to train FSD’s neural networks to better estimate distance and object volume in that blind zone. Essentially, Tesla is teaching the car to infer accurate obstacles in the void where its production cameras can’t see, ensuring it won’t clip a fender when navigating tight spaces.
The Banish Bottleneck
The urgency for this testing is double-natured. Banish is an essential feature for Robotaxi and FSD Unsupervised, and the Cybertruck is the only vehicle in the fleet that lacks Actually Smart Summon.
While the rest of the fleet is enjoying the fruits of FSD v14, Actually Smart Summon currently runs on an older legacy tech stack that hasn’t been migrated to the unified end-to-end neural network. This legacy code relies heavily on Autopilot code, which was never ported over to the Cybertruck.
Meanwhile, Banish is the feature where your vehicle can drop you off at the door, and then autonomously roam the parking lot to find a spot on its own. This feature, which still hasn’t arrived despite promises of it in FSD v12.5, requires your vehicle to navigate complex, unpredictable traffic in a parking lot without a human in the driver’s seat. No matter the vehicle, the margin for error is zero.
Tesla must have millimeter-perfect precision on its vision models in the forward blind spots to ensure that Banish or Smart Summon doesn’t cause an accident.
A Glimpse of Future Hardware?
While these brackets in the pictured vehicles are clearly temporary, they still raise an interesting question about the future. Will Tesla add additional cameras in future hardware iterations, or will they be able to solve the blind spot via software?
Competitors like Lucid and Rivian utilize additional cameras, radar, or ultrasonic sensors to handle these specific blind spots. While Tesla removed their ultrasonic sensors in the past to save on cost and complexity, as well as to reduce sensor confusion, relying entirely on high-mount cameras has proven to be an ongoing challenge.
Before Tesla added a front bumper camera, many were asking for Tesla to add cameras at the corners of the front bumper. These cameras would not only eliminate the front blind spot but also allow the vehicle to see around intersections with obstructions much earlier. Today, the vehicle needs to move forward enough for the B-pillar cameras to have a clear view of an intersection.
However, it’s highly unlikely that we will see a retrofit or sudden mid-cycle refresh adding these cameras to any of Tesla’s vehicles. Instead, this is almost certainly a data-gathering exercise to strengthen the software for existing hardware.
