AI boosts advances of Robots

 

The New Titans Behind Figure

By Jeff Brown

Held in June 2015, the DARPA robotics challenge was a sight to see. The goal was to bring together the most advanced robotics technologies in the world, capable of operating on the ground and performing complex tasks in (staged) dangerous environments. It was a useful exercise, which highlighted some of the applications for an intelligent humanoid or ground robot for humanitarian, defense, or industrial environments. For such a serious event, the results were quite comical. Highlights From the 2015 DARPA Robotics Challenge Robot Bloopers | Source: IEEE Spectrum on YouTube No More Robot Bloopers DARPA stands the Defense Advanced Research Projects Agency. It’s the research and development arm of the U.S. Department of Defense, tasked with staying on top of emerging technologies… and how they can be used to assist U.S. military. DARPA’s robotics challenge was eight and a half years ago. Here’s what the most advanced technology looks like as of this February: Tesla’s Optimus Going for a Walk Optimus Getting Its Daily Steps In | Source: @Tesla_Optimus on X DARPA’s robotics challenge provides great context for how far the technology has come.  In comparison, the technology from 2015 was laughable, and yet it was the best available at the time. Now, Optimus is capable of a remarkably stable gait, can change directions easily, and can even walk at speed that is faster than some of us humans who may be advanced in age. And as we saw in Outer Limits — The Race to Catch Optimus, Optimus is even capable of fine motor skills, like the kind required to fold clothes. What’s most remarkable about Optimus though is not what it can do today, but the fact that it didn’t exist two and a half years ago. It was August of 2021 when Elon Musk announced his plans for Optimus. I covered the announcement at the time, and longtime readers probably remember that event, as we all watched a human — dressed as a robot — trot out on stage and start dancing. Musk and his robotics team were laughed at. The industry mistakenly misinterpreted the humorous press event to mean that Tesla was years behind in robotics technology and wasn’t a competitive threat. They’re not thinking that anymore. Figure’s Giant Backers “We shouldn’t be distracted by all the humor,” I wrote back in 2021 upon Musk’s humorous Optimus announcement. “This is no joke — Tesla will deliver.” Tesla delivered. And so did another company Outer Limits readers know well. Figure AI, a promising competitor to Tesla’s Optimus, just raised $675 million at a $2.6 billion valuation. I wrote about a rumored raise for Figure AI last month in The Race to Catch Optimus. At the time, it was anticipated that $500 million would be raised at a $2.4 billion valuation. Clearly, the stakes increased, bringing in an additional $175 million at an even higher valuation.  But for a hot robotics/AI startup like Figure AI, it’s not the amount, or the valuation, that’s interesting… It’s who was behind the deal. Jeff Bezos invested through his family office Bezos Expeditions, in addition to Intel Capital, Microsoft, OpenAI, NVIDIA, LG Innotek, and Samsung, among others. Just look at those names: Jeff Bezos (Amazon), Microsoft, Intel, NVIDIA, and OpenAI. Amazon and Microsoft compete ferociously against each other as cloud service providers.  NVIDIA has been the biggest thorn in Intel’s side, completely dominating the business for data center semiconductors and becoming the most valuable semiconductor company in the world. And yet they all got together to invest in Figure AI. They’ve backed the company that they think has the best shot at competing with Tesla’s Optimus. No other humanoid robotics company has received this much “attention” to date. The reality is that they are running scared. Desperate to catch up. Musk and his robotics team’s progress has been nothing short of stunning since August 2021. And each month, Optimus is capable of demonstrating even more impressive feats. The pace is accelerating. 17%... and Quickening… The team at Figure may not be moving as quickly as the team at Tesla, but the progress has still been impressive. In January, we saw Figure 01 learn to make a cup of coffee using a popular coffee machine. Figure 01 Makes Coffee | Source: Figure AI on YouTube It’s obviously a very simple task, which could have been programmed into Figure 01, but that’s not what was demonstrated. In order to learn how to use the machine, Figure 01 watched a human make coffee 50 times in a row. That video input was the training set for Figure 01’s artificial intelligence to learn from. And after seven hours of practice and iteration, Figure 01 had mastered the task. This kind of learning is enabled by a multimodal approach to artificial intelligence. Multimodal AI means that an AI can learn from text, pictures, audio, data, and video — including any of the aforementioned categories perceived in real-time. And once trained, the AI can infer the correct actions to take in real-time. This is the kind of technology that OpenAI has been developing, which is why the investment from Microsoft and OpenAI makes so much sense. And just a few days ago, Figure demonstrated a real world application for Figure. It was a staged industrial setting where Figure 01 was tasked with picking up a bin and transferring to another location. Figure 01 Bin Transfer | Source: Figure AI It’s not nearly as impressive as what Optimus is doing, but like Optimus, Figure 01 has been developed in less than three years. The above demo is currently working at about 17% the speed of a human equivalent. While Figure 01 is still tethered, it demonstrates stable autonomous navigation, and the ability to lift and manipulate a bin. With the help of OpenAI and NVIDIA, Figure 01’s autonomous capabilities — as well as multimodal learning — will improve quickly. And I’m sure that Microsoft will provide as much computer processing services to Figure AI that it demands… in order to accelerate learning. Who’s Next? This is a multi-trillion dollar market in which Tesla is in the lead. So many questioned Musk’s strategy as a “car company” to get “distracted” with this side project to make robots. They’ve completely missed the point. Tesla has long been the world’s most successful artificial intelligence (AI) company. It just so happens that its first product was an autonomous electric vehicle. And having almost entirely solved the incredibly complex challenge of fully autonomous driving — developed on deep neural network technology — and mastered advanced electronics manufacturing at scale… it has an arguably unbeatable competitive advance when it comes to humanoid robots. The reality is, Tesla doesn’t need anyone’s help… Which is why other industry giants and billionaires are scrambling to back the next best thing.  I’m excited to see who’s next.