A well-funded startup is reporting a breakthrough to help humanoid robots to walk into any home and start helping immediately.
1X Technologies just introduced its new Redwood AI model powering its NEO humanoid robots. They’re racing to scale millions of AI-powered robots into homes amid rapid-fire development in the humanoid robotics space.
As its competitor Figure AI reports significant progress toward general-purpose robotics, Tesla’s Optimus is marching forward amid leadership changes and leading Chinese robotics firm Unitree is on the verge of going public.
Redwood AI
1X Founder and CEO Bernt Bornich says, for the first time, “NEO moves, understand, and interacts seamlessly in complex human environments.”
“Built to learn from real-world experiences, Redwood empowers NEO to adapt gracefully, turning failures into insights with hardware designed for compliance, safety, and resilience,” he posted on the X social media platform.
The 160-million-parameter transformer model combines vision, touch, and body movement data. It’s trained on a mix of teleoperated and real-world actions from its NEO and EVE humanoid robots. It runs full onboard, with no cloud connectivity required.
1X, which was founded in 2014 in Norway and is now headquartered in Sunnyvale, shared footage shot in Redwood, Calif., to showcase the new capabilities. According to the startup, the Redwood neural network enables the NEO Gamma robot in the demo to pick up objects it’s never seen before, even in new places and messy environments.
The model unlocks humanlike behaviors. While most robots treat walking and using their limbs as two separate things, Redwood controls everything at once. NEO robots now move their whole bodies to get things done similar to humans and decide on their own whether to use their left or right hand. Of course, NEO robots are ambidextrous by design so it’s based on logic rather than hardwired into their systems.
1X says the model is perfect for households because it works even if there’s no Internet connection. However, at this point, Redwood requires cloud connectivity to process verbal commands, but not for action execution.
The startup says the model is in its early development but improving thanks to real-world data collection. The AI also works on 1X’s legacy humanoid robot, EVE, which was introduced in 2018 and went viral last year thanks to popular streamer Kai Cenat.
1X has begun testing its humanoids in rea homes, starting with 1X employees who’ve shared clips of the robots performing household chores on social media. The startup plans to scale production of its next-generation robots in 2026 with a goal of manufacturing 100,000 units by 2018 and scaling into the millions from there.
1X has differentiated itself in the humanoid robotics race by targeting home deployment first rather than industrial applications.
Figure AI Scales Helix
Its competition includes Silicon Valley-based Figure AI, which is reporting fast progress with its self-developed Helix AI system.
Figure has two major customers: BMW and an unnamed logistics company. The startup plans to start home testing later in 2025.
In a new hourlong demo, one of the robotics firm’s flagship Figure 02 humanoids performs sorting tasks, adapting in real time. On social media, Figure founder and CEO Brett Adcock noted that the robot makes some mistakes in the footage but those are going down as Figure’s humanoids collect data from real-world actions.
Increasing training data from 10 to 60 hours boosted barcode success from 88 percent to 94 percent and improved handling speed by 58 percent. Helix learns subtle actions like padding down wrinkled plastic to flatten a barcode from demonstration data.
The system keeps a short-term memory of what it’s seen recently to avoid repeating actions. It remembers recent body positions to maintain fluid continuous motion. This enables on-the-fly corrections when a grasp fails or a package shifts.
Figure robots use a single neural to control tasks from just camera input. Individual robots can share what they learn with their whole fleet, like an AI robot hivemind.
During a recent interview with Bloomberg Technology, Adcock didn’t confirm, but also didn’t push back on, reports that Figure is raising $1.5 billion for a nearly $40 billion valuation. The startup aims to manufacture 100,000 of its next-generation humanoids in the next four years.
Unitree Heads Toward IPO
Meanwhile in China, Unitree Robotics is laying the groundwork for public listing.
Founded in 2016, Unitree is experiencing a significant growth spurt as its quadrupedal and humanoid robots become ubiquitous. The company recently opened its new 10,000-square-meter factory in Hangzhou near its corporate headquarters and is building out branch locations.
Going public could give Unitree access to lucrative contracts from large enterprises and governments that often prefer or require public listings for perceive transparency and financial stability.
According to the South China Morning Post, Unitree restructured from a limited liability company to a joint stock limited company in late May 2025. The conversion, which means Unitree can issue and transfer shares, is widely interpreted as positioning the company for an IPO in mainland China or Hong Kong.
Unitree is reportedly raising funding for a valuation of up to $2 billion. Analysts note that Unitree is preparing to mass produce its humanoids in 2025, which requires a capital-intensive scale-up.
If Unitree goes public, it would become the second Chinese robotics firm focused on humanoids to do so. Shenzhen-based UBTECH Robotics debuted on the Hong Kong Stock Exchange back in December 2023 with its Walker series leading the company’s humanoid lineup.
Optimus Lead Departs Tesla
Back in the US, Tesla just lost a key figure in its Optimus humanoid robot project.
Milan Kovak, the EV maker’s vice president of engineering and head of its Tesla Bot program, has stepped down, citing personal reasons. On X, he emphasized he’s leaving to spend more time with his family abroad. He also reaffirmed his ongoing support for Elon Musk and Tesla while insisting that it would have no impact on development timelines.
“I want to thank @elonmusk from the bottom of my heart for his trust and teachings over the decade we’ve worked together,” he posted on X. “Elon, you’ve taught me to discern signal from noise, hardcore resilience, and many fundamental principles of engineering. I am forever grateful.”
According to the Wall Street Journal, Kovac began at Tesla in 2016 on its Autopilot team and took over Optimus in 2022. Reports indicate that Ashok Elluswamy, Tesla’s Autopilot lead, is stepping in to oversee Optimus.
Kovac is leaving at a critical juncture as Tesla aims to manufacture at least 5,000 Optimus humanoids in 2025. Musk has said he believes the robot will become Tesla’s most popular offering, by far. Shortly before Kovac departed, he unveiled a new breakthrough that unlocked the ability for Optimus humanoids to learn complex tasks directly from Internet videos of people performing them.