The use of robots on construction sites has failed to reach its potential, says a University of Toronto researcher, but with more research, a fully autonomous, high-level mobile robot could be a step closer to achieving it.
Most of the current so-called robots that patrol construction sites should be more accurately referred to as tools that repeat some pre-programmed task, says Kim, an assistant professor in the University of Texas College of Applied Science and Engineering.
Aside from a few success stories, what is lacking is full robotic automation and digitization that uses human-level visual artificial intelligence (AI) to fully understand construction sites where they are deployed.
To reach the high level of visual artificial intelligence that will run bots in websites millions of images are needed but for a variety of reasons, getting that number is impractical. What Kim and his team are proposing are two new technologies: compiling virtual construction images, and generating building images on a miniature scale.
“As we’re developing new forms of construction robots, the hardware part has come a big step up in, say, Spot from Boston Dynamics, but the software development, that AI part, still has a long way to go,” Kim said.
The problem is that we lack training data for building scenes. DNN, Deep Neural Network, the core engine of visual artificial intelligence, is a supervised model, which naturally becomes greedy for data. To develop a well-trained AI… we need an enormous number of well-trained, diverse images for construction scenes.”
Kim’s research program was one of 251 undergraduate initiatives announced as recipients of a total of $64 million in funding from Innovation Canada’s John R. Evans Leadership Fund in September.
His project summary, submitted to the Innovation Center, stated that “Robotic solutions with enhanced AI will collaborate with field workers safely, improving productivity and profitability while offsetting a growing labor shortage. The proposed research project is essential to realizing this vision, providing models An improved field-applicable DNN – a critical next step in the development of autonomous construction robots.”
The robots will collect, analyze and document site information, allowing for the creation of live double digital models of ongoing construction sites.
Kim explained that superimposing the images to be developed into visual AI is required, first because it is difficult to collect data in person.
Surveillance cameras and drones have blockages, are very expensive—Kim mentioned $2 to $10 an image—and present other problems.
Collecting a million photos can take a long time and there are many problematic regulations and confidentiality issues.
Marketing and sharing data in a competitive build environment is another issue.
Work is moving quickly in Kim’s U of T lab, where the team uses five Tensor processing units and Google Cloud software. More computational resources are required.
“We’ve been fully focused on developing simulation software that can automatically superimpose unreal but real-looking building images, and a few weeks ago, we started actively creating a million construction training images. This is exciting news to me, as far as I know, we haven’t had the opportunity from Prior to using a million training images in DNN construct training.
Assembly steps include creating a 3D human model, followed by entering motion capture data for the workers; Create a 3D construction worker avatar by mapping the 2D or 3D clothing map onto the 3D human model; Randomly set shooting conditions, including camera distance and lighting conditions; Synthesize and create construction images or videos by superimposing a virtual construction worker avatar on 3D construction backgrounds.
Next comes a prototype of a fully autonomous mobile robot for the building’s digital twin that deploys higher-order DNN models.
Construction robots will need to be able to monitor and analyze the location, moving speed and direction, posture, proximity, and other factors that capture the presence of construction workers.
“It remains unclear how effective artificial imagery is at training visual AI models of a building scene, which is highly dynamic and unstructured.” Kim said. “We may or may not need our own unique solution.”
For the last step, Kim will need partners in the private sector – he is looking for an innovative construction company that would financially support the search.
Follow the author on Twitter @DonWall_DCN
#prof #seeking #high #level #construction #robots #artificial #intelligence