VCs like Khosla and A16z are backing next-generation robotics that leverage AI

VCs like Khosla and A16z are backing next-generation robotics that leverage AI
VCs like Khosla and A16z are backing next-generation robotics that leverage AI

The robotics industry is undergoing a profound transformation thanks to a new AI offshoot called “spatial intelligence,” and venture capital giants such as Khosla Ventures, Andreessen Horowitz and General Catalyst have recently backed numerous startups building robots capable of advanced reasoning and visual data processing.

Spatial intelligence enables robots to navigate and interact with their environment more efficiently. This means robots can perform more complex tasks with greater precision and adaptability.

Although robots have been in use for decades, for the most part they are not particularly “intelligent.” They excel at performing repetitive tasks in controlled environments without surprises.

The next generation of robots will be able to sense, see, make decisions and act based on their goals and their environment.

Take, for example, screwing a lid onto a jar. With traditional robotics, these tasks might require multiple machines, each individually programmed to pick up a jar, put a lid on it, and then screw it tightly. The robot might have trouble completing the task if there is something wrong with the lid, the jar, or the placement.

With spatial intelligence, on the other hand, a user could tell a single machine to screw a lid onto the jar in plain language rather than code. The machine, in turn, would interpret that request, teach itself how to screw a lid onto the jar by watching a person, YouTube, or another robot, and then complete the task—and repeat this process for any further requests from a user.

“The introduction of the LLM in recent years has been a major breakthrough in robot education and the market is huge,” said Howard Morgan, chairman of B Capital Group.

He explained that as spatial intelligence advances, robots will be able to teach themselves to perform a variety of tasks, meaning the engineer won’t have to build a highly specialized machine that can only do one thing.

“For the first time, you can deal with the ambiguity and randomness of life that robots aren’t normally well suited to,” said Max Rimple, an investor at General Catalyst. “AI is this really huge breakthrough where robots can be used to do tasks that people never thought robots could do and be useful for.”

Big world, big data

The technology is far from perfect, and several investors told Business Insider that the biggest challenge right now is getting enough data to give robots spatial intelligence. Unlike large-scale language models (LLMs) like OpenAI’s Chat-GPT, which are trained on large amounts of text-based data to spit out accurate and human-sounding answers to written or spoken questions, there is far less data to train robots’ spatial intelligence.

“The biggest bottleneck in building a base model robotics company similar to OpenAI or something similar is that it’s very difficult to achieve internet-scale data for robotics,” said Erin Price-Wright, an investor on a16z’s American Dynamism team, which focuses on AI for the physical world.

“Data is much scarcer,” she added. “The largest dataset ever published is tiny compared to the ability to search the entire internet.”

Still, investors say spatial intelligence represents a quantum leap for the robotics industry, which is already benefiting from an AI boom. Early- and growth-stage robotics startups have raised more than $4.2 billion so far this year, according to Crunchbase, though some have kept details private, including Skild, which was in talks this spring to raise $300 million from Lightspeed Venture Partners and Coatue Management, and a new startup founded by prolific Stanford researcher Fei-Fei Li, widely known as the “godmother of AI.”

Kanu Gulati, partner at Khosla Ventures, said she is seeing more startups springing up that are working in different, creative ways to solve spatial intelligence data problems, which is good for the race to create a viable end-to-end model.

“Most companies need to have a certain number of robots collecting real data in a cost-effective way. That creates a flywheel,” she said.

With this in mind, Gulati and Khosla have invested in startups with various applications across different industries that are working on developing spatially intelligent robots. Their portfolio includes logistics startups Waabi – which offers autonomous truck rides – and Vayu – last-mile delivery robots that use bike lanes – as well as climate technology startup Zorbi, which builds autonomous greenhouses, and FieldAI, which develops hardware-agnostic software for any kind of robot.

“Over time, solving the data challenge will mean getting a working end-to-end model and not just individual components,” Gulati said.

Lior Susan, CEO and founder of Eclipse Ventures, believes that the effort put into developing AI with spatial intelligence will lead to greater returns in the future.

“You’re seeing this boom in generative AI startups that are actually buying a lot of computers but don’t have a business model yet,” Susan said. “The people who are connecting unique data sets from the physical world using AI – not scouring the internet – are the people who are going to make money, in my opinion.”

The big technology companies are also on the rise

Larger technology companies are also gearing up to join the race for spatial intelligence in robotics – both as developers and investors. OpenAI is relaunching its robotics research group and is currently hiring research engineers with a focus on robotics. Hugging Face, a well-known platform for sharing AI models and datasets, also launched a robotics program this spring.

Physical Intelligence, which builds general-purpose, AI-powered robots, raised funding in March from OpenAI, Khosla, Lux Capital, Sequoia Capital and Thrive Capital. Figure, which also builds general-purpose robots, raised $675 million this spring from Microsoft, OpenAI, Nvidia and Jeff Bezos. Through its venture capital arm, Nvidia also led the May funding round for Carbon Robotics, which uses AI lasers to kill weeds on farms.

“The next generation of AI needs to be physical,” said Jensen Huang, CEO of Nvidia, at the Computex conference in Taipei last month. “Most of today’s AIs don’t understand the laws of physics. In order for us to be able to create images, videos, 3D graphics and many physical phenomena, we need AIs that are physical and understand the laws of physics.”

Leave a Reply

Your email address will not be published. Required fields are marked *