Request A Technical Demo or Trial
Let's explore if we can help accelerate your perception development and deployment.
Thank you! We'll be in touch soon.
Oops! Something went wrong while submitting the form.
Let's explore if we can help accelerate your perception development and deployment.
When a robot relies on sensors for safe operation and performance, the calibration of its sensors is of the utmost importance. In order to produce a sensor-reliant product that scales, the calibration process should be quick and simple enough that an untrained operator can perform it. In this case study, we outline the previous calibration process of a mobile robotics company and how it was reimagined to save time, reduce labor costs, and meet their production goals.
The company’s previous calibration procedures took three hours to process a multi-camera data set with an off-the-shelf open source tool and required a full-time robotics engineer to calibrate each system. Given these constraints, the company was able to produce a maximum output of 693 robots per year.
With an increased demand for their product, the company needed not only to reduce the amount of time spent calibrating each system, but also needed a simplified calibration process that an unskilled operator could successfully perform on a production line in order to meet their new annual production goal of 2,400 robots.
After uncovering the company’s approach to calibration, Tangram Vision’s team of perception engineers set out to revamp and replace it. As a first step, they implemented advanced filtering to ensure the quality of inbound datasets and reject frames that would lead to poor or failed calibrations. Tangram Vision’s advanced data filtering rejected thousands of unnecessary frames, and selected a small subset of golden frames required for calibration success. As a result, Tangram Vision turned what was previously a three-hour calibration data processing step into one that takes only three minutes and 28 seconds; thereby reducing the time to process the calibration data by 98 percent.
For the last piece of the puzzle, Tangram Vision’s team worked with the company to identify verifiable precision and accuracy thresholds in order to create a simplified calibration process for their production line that would produce a foolproof pass/fail result. By adding the new quality thresholds to Tangram Vision’s calibration platform, the company is able to leverage a graphical user interface to provide unskilled operators with one-touch data capture and a straightforward pass/fail notification at the end of the calibration process.
Tangram Vision’s platform provides detailed metrics for engineers (shown left) and simple pass/fail results for the production line and customers (shown right).
While Tangram Vision simplified the results the production line operators, detailed calibration metrics and history for each robot remain fully available to engineers for future system troubleshooting. The metrics for each sensor in their fleet include extrinsic calibration values plus covariance, intrinsic values plus covariance, stereo pair synchronization records, and binned 10x10 camera coverage charts to ensure enough data was captured across each camera. The ability to access historical calibration metrics will allow the company to perform in-field calibration checks and root cause analysis over each robot’s service life.
By implementing Tangram Vision's calibration system, the company is able to increase their production output by 246%. The reduced calibration time and simplified process allowed the company to transition from a skilled robotics engineer to an hourly production line worker, not only resulting in cost savings but also enabling the company to reach its desired annual output of 2,400 robots.
