Deepen AI Launches Self-serve Targetless Calibration API for Autonomous Vehicles and Robots
News related to:Deepen AI · 2 min read
Deepen AI, a company that has been calibrating production sensor rigs for automotive and robotics teams since 2017, has launched a new self-serve Targetless Calibration API. This tool allows teams to calibrate lidar-camera rigs from recordings, eliminating the need for dedicated calibration sessions, targets, or site visits.
The new API, which is available today as a web app and an API, enables engineers to calibrate every sensor in a recording with just a short drive through a mostly static scene, as little as 30 seconds. The engineers can also provide a rough position for each camera, taken from CAD, a URDF, or the recording's own transforms. A single run solves all sensor pairs jointly, ensuring that errors do not stack up and the whole rig comes back in one consistent model.
Deepen AI's Targetless Calibration API supports a wide range of existing formats, including ROS1 .bag, ROS2 .db3 and .mcap files, and even a zipped ROS 2 bag folder. The service is designed to work with various lidar models, such as Velodyne, Ouster, Hesai, and RoboSense, among others. Every job returns per-sensor confidence scores, quality flags, error statistics, and a visual overlay report showing the seed-versus-recovered delta, allowing reviewers to judge the result.
Teams can try the service for free before paying. Every run shows a free before-and-after preview of the lidar on each camera, with per-pair confidence. If the team chooses to download the result, they pay $150, which is 90% off the $1,500 list price, through December 31, 2026, with no monthly commitment. During the same quarter, Deepen AI's calibration engineers will support any team that wants to try the product, from preparing a recording and setting camera positions to running the first calibration together.
The service supports a variety of applications, including passenger and commercial vehicles, delivery and mobile robots, construction, mining, and agricultural machines, drones, ships, and tracked platforms. The company claims that independent 30-second recordings on their test vehicle agree to within 0.1° angular, 1 ms temporal, and 2 cm translation.
For teams with data-residency or export-control requirements, a customer-hosted deployment runs the same engine inside the customer's own environment. An open-source Data Checker CLI (deepen-bag-check, Apache 2.0) checks a recording locally, ensuring that no upload or account is necessary.