In 2017, Fei Fei Li, Stanford’s renowned computer scientist who is considered the Godmother of AI in the industry, spoke with Melinda Gates in Wired magazine about ‘liberating AI from guys with hoodies’.
In April 2024, Li gave a talk at TED in Vancouver and said, ‘the cutting edge of research involved algorithms that could plausibly extrapolate what images and text would look like in three-dimensional environments and act upon those predictions using a concept called spatial intelligence.’
In August 2023, TechCrunch reported that Li’s new startup, World Labs, had raised $100 M in funding. The company is valued at $1 Billion.
World Labs is developing a spatial intelligence model that can accurately estimate the three-dimensional physicality of real-world objects and environments. This will enable detailed digital replicas without extensive data collection. Sources say the startup’s artificial intelligence (AI) will also be capable of advanced reasoning.
According to Johannes Maunz, VP of AI at Hexagon, the growing interest in spatial intelligence models and related startups is a positive trend for the broader industry.
“This focus on spatial intelligence reflects a broader recognition of the importance of how we interpret physical spaces in a digital world,” said Maunz. “Investment and research help to accelerate innovation, leads to more sophisticated algorithms, improved hardware and more applications.”
Maunz believes that could increase the accessibility of spatial intelligence technologies, which benefit the industry at scale. “We expect that increased industry awareness of spatial AI models will have a positive impact.”
Spatial intelligence and digital twins
“As advancements in AI technology continue, the next big step is to bring spatial intelligence and digital twins together,” he said. Maunz points to the company’s digital twin in Klagenfurt, Austria, as an example of combining spatial intelligence and digital twin technology.
“Through spatial intelligence technology, we combined reality capture data from the city with AI-enabled software to create a digital twin of Klagenfurt in 3DT,” said Maunz. “The digital twin provides measured information about properties and how they are composed. In detail, with spatial intelligence, we separate the area into classes such as grass, water, building, road to provide the exact details.”
From those details, Maunz says they produced a 3D photorealistic model of the entire city. “This information is pivotal to simulate solar panels, discover heat islands, and know the level of impervious areas across the entire city and with that data, Klagenfurt can simulate specific actions upfront, implement and measure them afterwards.”
Sensor data
Hexagon’s sensors have created more than 150 petabytes of privately owned data. Maunz says that data can be used to train AI models based on data captured from the real world.
One petabyte of storage is equivalent to 11,000 4 K movies, and 150 petabytes would be 214,041.096 years of playing video games.
“With the rise of computer vision foundation models, the results of spatial intelligence models will become more powerful, versatile and reliable,” he added.
Geospatial intelligence
Data scientist Vkhyat Chaudhry is the co-founder and CTO at Buzz Solutions. He takes spatial intelligence in another direction, looking at geospatial intelligence and real-world applications being enabled by AI. Buzz Solutions focuses on managing and monitoring deforestation and critical infrastructure inspections in sustainability, agriculture, and environmental protection.
Sustainability, agriculture and natural disasters
“Geospatial AI can be used to analyze and interpret spatial and geographic visual data either captured through satellite or aerial (drones, fixed wing, helicopters),” said Chaudhry. “Emerging use cases include applications for environmental sustainability.”
Chaudhry says that visual data from satellites and aerial vehicles can be analyzed using computer vision-based AI algorithms to detect changes in forest cover, pollution, carbon, methane, and other greenhouse gas emissions, sea level changes, and other environmental impacts.
“Geospatial AI can applied to agriculture as well,” he said. “Multi-spectral visual data analysis can be used to detect crop health, soil texture and conditions, predict crop yield, moisture and nutrient content for enhancing agriculture and farming precision.”
Chaudhry says that looking to prediction and detection of natural disasters, he sees Geospatial AI playing a role. “Aerial imagery can be analyzed by machine vision algorithms to detect disasters such as wildfires, landslides, etc., and can also support disaster prediction, response and recovery.”
Inspecting critical infrastructure
Geospatial and aerial imagery capture for power and energy infrastructure can help monitoring and maintenance.
“Computer vision algorithms can detect various power grid components, equipment anomalies and defects using visual RGB and thermal imagery,” said Chaudhry. “Buzz Solutions provides the monitoring and the visual data of power grid infrastructure, so this helps in more efficient, safer and faster inspections and maintenance of the power grid, hence preventing infrastructure failures that could lead to massive power outages, blackouts and even wildfires.”
Chaudhry also says that satellite, aerial and LiDAR imagery are analyzed using advanced computer vision algorithms to detect vegetation growing near power lines and power grid infrastructure.
“Understanding the areas of excessive growth and growth patterns of vegetation around this highly energized infrastructure is important to provide effective vegetation management,” he said. “Multi-sensor and multi-spectral data, including near-infrared (NIR), Normalized Difference Vegetation Index (NDVI) and other techniques combined with advanced computer vision algorithms helps in detecting vegetation and predicting vegetation growth over time.”
One advantage of managing vegetation around infrastructure assets for Buzz Solutions’ utility customers is preventing the possibility of wildfires, which cause millions of dollars in damage annually.
Spatial computing on the horizon
Looking towards the next three to five years, with the rise of Chat GPT, Maunz believes there has been an increase in general acceptance of AI-based solutions.
“Rapid progress is also being made in researching spatial AI. Over the next 3-5 years, I expect we will see more real-life applications emerge from conceptual research. I expect spatial computing to become increasingly integrated into our daily lives and work processes,” said Maunz.
