Answers to numerous questions (Are there three points of contact with the person on the ladder? Are they? To use the ladder as stelots to rotate?) Are combined to determine that the ladder is being used safely in the picture. “There are a dozen layers of inquiries in our system just to reach this answer,” says Lorenzo. The drone has not released its data publicly to review, but says they hope they will audit their procedures freely by protective experts..
5 % missing
Chen Feng says the use of vision language for construction AI shows promise, but there are “some beautiful basic issues” to solve, including the problem of deception and age cases, the irregular risks for which VLM has not trained. He leads the lead AI4CE Lab of New York UniversityWhich develops technologies to understand 3D mapping and scene in construction robotics and other areas. “Fifty-five percent is encouraging-but how will we fix the remaining 5 %?” He asks for the success rate of Safety AI. Feng indicates a dissertation of 2024 that “saysEyes clipped off.? “-Nyu is written by PhD student Shengbing Tong, and in collaboration with Ai Luminary Yann Lecun, which notes” organized shortcomings “in VLMS.” To identify the object, they can well reach human level performance. He noted that VLMS has struggled to translate a 3D scene structure with 2D imagery, not aware of the good situation in the argument about local relations, and often lacks “general understanding” about visual scenes.
Lorenzo acknowledged that LLM has some “big flaws” and he struggles with local reasoning. Therefore, Safety AI has also used some old methods of learning machine to help create a local model of construction locations. These methods include distributing images into important components and photographer, which is a established technique to create a 3D digital model from a 2D image. Safety AI has also received heavy training 10 different problems areasThe most common violations, including the use of the ladder.
Nevertheless, Lorenzo acknowledged that there were cases that would fail to recognize LLM. But he notes that for more working safety managers, who are often responsible for more than 15 sites simultaneously, an additional set of digital “eyes” is still an improvement.
Aaron Tan, a concrete project manager based in San Francisco Bay Area, says that a device like Safety AI can be helpful for these overturned safety managers, who can personally receive email warnings instead of making a two -hour drive to go to a site. And if the software can show that it is helping to keep people safe, he thinks that workers will eventually embrace it.
However, Tan notes that workers also fear that such tools will be “bossware” that will be used to trouble them. He says, “In my last company, we imposed a security system (AS) camera. And the boys didn’t like it.” “They were like, ‘oh, older brother. You guys are always watching me – I have no secret.”
The meaning of the old is not obsolete
Ezak Paz, CEO of a company based in Jerusalem, has considered the inclusion of VLM under the name of Safe Guard AI, but he is trapped with an older machine learning paradim because he considers it more reliable. He says, “Old computer vision” based on machine learning is still better, as it is hybrid between the machine and is human intervention in dealing with deviation. ” To train the algorithm in a new category of danger, his team collects a large amount of specific risk -related footage and then trimmed the false and false negatives and improves the algorithm. Paz says the process can take more than six months to six months. With the completion of training, the Safe Guard makes a risk diagnosis to indicate potential risks on the AI ​​site. It can “see” the site in real time by accessing footage from a camera connected to any nearby Internet. Then it uses an AI agent to forward what to do with the site administrators’ mobile devices. Paz refuses to give a precise price tag, but says its product is cheaper for builders only at the “mid -market” level and above, especially those who manage to multiple sites. The device is used in about 3,500 sites in Israel, the United States and Brazil.