Computer Vision in Construction
On March 3, prior to the lockdown, I visited a tech conference at UC Berkeley, called TC Sessions: Robotics + AI 2020. It was a showcase of innovations that both startups and already established AI companies had created. Out of all of the creations shown that day, I found a start-up company called BUILT Robotics the most promising.
BUILT Robotics functions as both a construction and AI company. Founded in 2016, the company has dedicated its efforts to training construction robots to act autonomously. There are currently three functioning vehicles that are modified to be unmanned. These vehicles have extra sensors, detecting video, audio, proximity, etc. The company currently has a bulldozer, excavator, and a compact track loader.
The actual AI behind the robots works by using a combination of algorithms. First of all, a geofencing algorithm is used to restrict the overall movement of the vehicles. This is done by establishing a system of coordinates and thus confining the robot to a specified work zone. An object detection algorithm (I presume a CNN but that is speculation), is used to detect key elements in construction. To provide safety on the work-site, the algorithm stops path movement if a foreign object, such as a human or animal is in the way.
Overall, this technology looks promising. Sadly, none of the algorithms are open source as the company is for-profit. We look forward to what BUILT Robotics will accomplish in the future.
