Skip to main content

Sensor calibration for autonomous vehicles: The measurement behind the measurement

An increasing number of autonomous vehicles are in operation on today’s job sites and roads without a human operator, enabling industries to work smarter — not harder. Robotaxis are driving us around while we nap, work, or scroll through social media on our phones. It’s critical to have complete confidence in the partner you choose to enable never-before seen levels of efficiencies within agriculture, construction, mining and transportation. 

Behind the scenes, sensor calibration makes the data reliable

Enabling autonomy requires a multitude of checks and balances built into the process. Some get more attention than others. The truth is, sensor calibration doesn’t always get the spotlight it deserves. And it should, since all AVs come with a suite of sensors — from LiDAR and radar to cameras, inertial measurement units and more. Without knowing their proper placement, the spatial data they collect can be misinterpreted, which can have catastrophic consequences. 

Consider how an uncalibrated sensor might signal to an AV that it’s centered in the proper lane, when it might actually be in danger of crossing the centerline. For times when GNSS is unavailable or degraded beyond centimeter level positioning, it’s imperative that sensor data is on point. That’s why the process to keep sensors exactly where they should be is automated, with built-in procedures to ensure nothing is left to chance. 

Sensor calibration is the process of determining the proper placement and orientation of sensors for optimal data consistency from one sensor to another. Consider all the sensors that made our partnership with IHI Corp. a reality. This Japanese heavy industry manufacturer is successfully using Trimble’s map-based localization and other sensor fusion technologies to extend operation of its autonomous container and haulage trucks fleet into indoor and other GNSS-denied environments. Proper sensor calibration ensures positioning accuracy and repeatability. 

Similarly, Dynapac’s compactor augmented with Trimble’s advanced positioning technology drives autonomously to maximize productivity. Without sensor calibration, it cannot provide the required accuracy needed to function. In passenger vehicles, failure to calibrate a sensor can result in faulty data that can disrupt the functionality of advanced driver assistance systems. For instance, an autonomous vehicle’s sensors could perceive a pedestrian as farther away than it actually is.

Positioning accuracy & safety depend on sensor calibration

Autonomous vehicles use perception sensors to observe the world and make decisions. And these decisions justify the overlap of safety measures to ensure that the data coming out of sensors can be trusted. The fact that so little discussion is given to sensor calibration is a testament to how seamlessly it runs in the background. However, that doesn’t make it any less critical, especially since common scenarios such as temperature, dust, wind and even vehicle vibration can cause movement in or reduce the integrity of sensors over time. Sensors may also accidentally get physically bumped out of alignment.

Achieving the level of accuracy required for a vehicle to safely operate autonomously in the real world can’t be done with a tape measure. It requires the sophisticated synchronization of data, optimization, and outlier detection algorithms. Ideally, these algorithms would run automatically in real-time, with no need for human intervention.

Identifying the same object in data from different sensors is a challenging problem as an object may look very different from sensor to sensor. Modern calibration algorithms are turning to artificial intelligence and machine learning to solve this matching challenge. Algorithms based on deep learning use data to predict what a pedestrian in a camera’s image will look like in a LiDAR point cloud. Using both real and simulated data provides a more robust solution that accounts for the infinite possibilities of inter-sensor data association.

Sensor calibration is not easy, and you cannot have autonomy without it. Automating the sensor calibration process avoids countless man-hours for operators, technicians, and engineers spent fiddling with expensive calibration targets. In the end, less time-consuming manual calibration means a more seamless autonomy solution, and more operation up-time in the field.