California-based startup Doxel developed an autonomous robot that captures 3D scans of construction sites and then sends that data to their deep learning algorithms to inspect the quality of the work installed and measure quantities in real-time.
“You can’t improve what you can’t measure. Without real-time visibility into quality and progress, managers simply can’t boost productivity. Our turnkey solution literally tracks progress for hundreds of thousands of line items in project budgets and schedules, comparing actual performance to original plans,” said CEO and Co-Founder Saurabh Ladha. “This is transformative for an entire capital project team. With Doxel’s system in place, project managers can react in minutes, not months, increasing productivity by fifty percent and bringing in projects twenty-five percent under budget.”
Using a variety of Quadro P5000 GPUs and GeForce GTX 1080 Ti GPUs with the the cuDNN-accelerated TensorFlow deep learning framework, the team trained their system to recognize different features of a construction project, like ductwork and wiring conduits, using 3D scan data.
“Construction projects involve millions of similar-looking components, packed tightly together in a dark environment. It’s a Molotov Cocktail of challenges for computer vision software,” said CTO and Co-Founder Robin Singh. “Usually, the amount of training data such an algorithm would need would be enormous, and even then, the results may not be reliable. With our proprietary 3D semantic algorithms, we’ve been able to get more reliable results with a fraction of the training data. Our algorithms recognize and contextualize objects not just based on color, but also based on shape, location, and size.”
On-board the autonomous robot is the NVIDIA Jetson platform which autonomously roams a construction site every day, both outdoors and indoors, with LIDAR and HD cameras. GPUs in the cloud then processes the visual data, inspects installation quality, and quantifies how much material has been installed correctly. Doxel’s cloud-based dashboard provides project managers with real-time feedback on productivity, as well as how actual costs and time spent are comparing to the original budget and schedule.
Kaiser Permanente used Doxel on a recent construction project in San Diego and was able to bring it in 11 percent under budget because of the software’s feedback. The company and its contractors were able to increase labor productivity by 38 percent.