Radically accelerate your roadmap with Tangram Vision's perception tools and infrastructure.
Table of Contents
When it comes to perception systems for robotics and automation, sensor calibration is absolutely critical — but often misunderstood. Despite its appearance as a behind-the-scenes infrastructure process, having a properly calibrated sensor array has far reaching impacts on the success of nearly all downstream sensor-driven applications from SLAM to machine learning pipelines, and more. Yet few robotics companies give calibration the attention it deserves until the impact of substandard calibration forces them to pay attention.
Why? There’s a common misconception that building calibration is a simple task, enabled by a plethora of legacy tools that make it a solved problem. However, the reality couldn’t be further from the truth.
At first glance, legacy calibration software might seem like an ideal solution for autonomy companies looking to tailor a calibration pipeline to their specific use case. It's accessible, modifiable, and, in theory, adaptable to a specific set of requirements. Upon closer look, it is clear this is not the case.
Legacy calibration tools are rarely robust enough for the sophisticated sensor arrays used in modern autonomy. Many of these tools were built in an era when robotics was still a hobbyist pursuit, and sensor arrays were relatively simple. As a result, those tools built a decade ago can’t be easily adapted to modern sensor suites that use multiple sensors of various modalities. They’re also not optimized for the realities of scaled deployment — namely, ease of use by non-engineers, and processing speeds that meet the demands of high throughput environments like production lines and in-field maintenance centers.
For companies that attempt to modify these tools to fit the needs of their sensor arrays, this lack of capability often necessitates significant modifications to the codebase, consuming substantial time and engineering resources.
Here at Tangram Vision, we speak with dozens of perception teams every year, and we often hear the same story over and over. It goes something like this: "We started with one perception engineer working part-time on calibration. After a year, we had made no progress, so we hired a dedicated calibration engineer to focus solely on the challenge. After two years, we still didn't have a workable system, so we hired two more engineers. It's been three years now, and we can finally calibrate our sensor array, but it takes hours to complete, and one of those engineers needs to run the process. Now we need to scale to grow, but calibration is holding us back. Help!"
For many of these companies, had they known how much time and money they were about to invest, they would have chosen not to build calibration in-house. But it can be difficult to understand the true cost of this decision. That cost covers more than just the engineering time spent trying to build calibration. It also includes lost revenue because of delayed time to market, and added engineering costs to cover product areas where engineering was pulled into calibration. To help you dig into these costs for your own company, we’ve put together a spreadsheet that’ll help your organization uncover the true cost of building calibration in-house. Just make a copy of this Google Sheet, and fill it in with actuals or estimates to gain a better understanding.
Developing a robust, reliable calibration pipeline isn't a task for a generalist. It requires the expertise and unique knowledge of a calibration specialist. However, in many companies, calibration is often assigned to engineers whose main job is not calibration, and whose first approach will be to try and adapt legacy software that will never be able to deliver a calibration pipeline for their sensor array.
This division of focus and false start can lead to less efficient calibration processes, subpar results, and ultimately, an unscalable perception system. It can also lead to the overextension of your engineers, negatively affecting their productivity in their primary roles.
Not all calibration pipelines are created equal. Those that are flawed can deliver faulty calibrations that appear to be correct from a data output perspective. Yet, in reality, the calibration values that are delivered are incorrect.
What happens when calibration values are incorrect? Critical processes like robot navigation and SLAM are misinformed by incorrect sensor data, leading to drift in trajectories that build over time to cause major errors. Safety processes like object detection and obstacle avoidance may operate outside the margin of error that they are intended for, thereby putting operators and others at risk. And inputs into other teams’ work, such as machine learning teams, are fed faulty sensor data that can throw off training, algorithms, and calculations, leading to wasted cycles and poor results.
With these challenges in mind, developing an in-house perception calibration pipeline frequently turns out to be a subpar approach. This is where Tangram Vision comes in.
Over the past three years, we’ve focused solely on creating a comprehensive, efficient, and fool-proof perception platform with an expertly-crafted calibration pipeline. Our team of specialists has built a process that addresses the unique challenges of modern perception system calibration, delivering quick and reliable results. And the results speak for themselves. Our partners benefit from highly accurate and precise calibration that can reduce the time spent on calibration processes by up to 98%.
The Tangram Vision calibration pipeline is not a side project or an additional task for our engineers — it’s our primary focus. This dedication and specialization allows us to provide a calibration solution that is not only quick and easy to use, but also highly accurate and reliable.
With Tangram Vision, you get the benefits of a specialist's expertise without the time and resource investment required for developing an in-house solution. You can focus on what you do best, confident in the knowledge that your perception system's calibration is in good hands.
While it may seem logical to develop a perception calibration pipeline in-house, the time, resource costs, and specialist knowledge required make it a challenging task. Partnerning with a dedicated, specialized provider like Tangram Vision is a cost-effective solution that ensures a reliable, high-quality calibration process. After all, for perception systems, reliability isn't a luxury — it's a necessity.
If you’d like to learn more about how we can solve calibration for your company, get in touch.