- Published on
K-Scale Labs: Betting on Open Source for the Future of Humanoid Robotics
- Authors
- Name
- The Editorial Team
- @humanoidsdaily
The Open Door: K-Scale Labs Challenges Humanoid Robotics Norms
Palo Alto-based K-Scale Labs, a Y Combinator Winter 2024 alumnus, is taking a decidedly different path in the burgeoning field of humanoid robotics. While many competitors operate under a veil of proprietary development, K-Scale is championing an open-source model for both its hardware and software. This approach, the company argues, is not just a philosophical stance but a pragmatic strategy to accelerate the arrival of general-purpose robots.
The company's mission, as stated on their website, is "to accelerate the timeline to a world with billions of general-purpose robots, by making them open-source and universally accessible." Founder and CEO Benjamin Bolte, a former AI researcher at Tesla and Meta, believes that the complexity of creating truly capable humanoid robots necessitates a collaborative, open approach. In a February 2025 interview with tech influencer Robert Scoble, Bolte remarked, "I want to make sure that that future is shared by everyone and is open-source and hackable... It's kind of a sad future to me if the only way to get a robot is someone that sells you a big black box."
This philosophy is perhaps best encapsulated in a tweet from Bolte: "Humanoid robots are essentially decentralized compute at scale. Doesn't make sense to be closed source, it is inherently a worse product."
Humanoid robots are essentially decentralized compute at scale. Doesn't make sense to be closed source, it is inherently a worse product
Attention Bittensor $TAO peeps ... Might have found the open humanoid hardware needed for a Humanoid Subnet. Looks like we won't have to pull a K2SO jailbreak on Optimus after all :)
The Great Debate: Open vs. Closed in Robotics
Bolte's assertion has not gone unchallenged. Sankaet Pathak, founder of competing humanoid robotics company Foundation, countered, "This is a bad take. Tesla cars are also distributed compute at scale. It’s not a better or worse product because of it being closed. What determines a better or worse product is the value it creates for its users. Open source vs closed is not the right debate. Build something people love."
This is a bad take. Tesla cars are also distributed compute at scale. It’s not a better or worse product because of it being closed. What determines a better or worse product is the value it creates for its users. Open source vs closed is not the right debate. Built something
Humanoid robots are essentially decentralized compute at scale. Doesn't make sense to be closed source, it is inherently a worse product
This exchange highlights a central debate in the rapidly advancing robotics industry: Will an open, collaborative ecosystem or a more traditional, closed-source approach ultimately lead to more capable and widely adopted humanoid robots? K-Scale argues that openness provides "reduced friction, greater collaboration, and increased visibility," outweighing monetary downsides and acting as a "forcing function for quality." They believe that for humanoid robots to achieve mass adoption, they must be "auditable, and broadly accessible."
From "Stompy" to K-Bot: K-Scale's Open Arsenal
K-Scale has been quick to translate its philosophy into tangible projects. Their early prototype, "Stompy" (officially Zeroth-01), is a 4-foot tall, 3D-printable humanoid whose CAD files, assembly guide, and bill of materials are publicly available, demonstrating a sub-$10,000 build cost.
Their flagship is the K-Bot Founder’s Edition, a 4'7" humanoid platform priced around $8,999, designed for developers and researchers. Alongside this, K-Scale launched the Zeroth Bot (Z-Bot) via Kickstarter in early 2025. This smaller, 1.5-foot humanoid, priced around $999, aims to make humanoid development more accessible to students and hobbyists. Both robots run on the K-Scale Operating System (KOS).
KOS, developed in Rust for real-time performance, is open-sourced under a permissive MIT license. It provides a hardware abstraction layer and gRPC APIs for high-level control, along with a Python library (PyKOS) for easier development. Complementing this is KOS-Sim, a simulation backend, and K-Sim, an open-source reinforcement learning library built on MuJoCo and JAX for training robot behaviors.
A key component of their software stack is the EdgeVLA (Vision-Language-Action) model. This multimodal AI, designed for on-device execution, allows robots to interpret voice commands and visual inputs to generate actions. The model and its training code have also been open-sourced.
"We want to have software problems instead of hardware problems, basically," Bolte explained to Scoble, emphasizing their strategy of starting with low-cost, accessible hardware and iterating on the software. "If you're designing a robot for RL (Reinforcement Learning), you can actually, like, make it quite a bit cheaper in some ways. You don't need to be as precise"
Building a Community, One Robot at a Time
K-Scale's open approach has fostered a significant developer community, with over 2,000 members on Discord. The company sees this community as crucial. "The nice thing about open source is that we get a lot of contributors," Bolte noted. He envisions a future where improvements are made in a distributed way, enhancing the safety and capability of all robots on the platform, similar to how Tesla updates its FSD software.
The company has received backing from Y Combinator and an early nod from OpenAI CEO Sam Altman, who reportedly pre-ordered a K-Bot and whose co-founded nonprofit donated $250,000 to K-Scale's open-source efforts.
The Path Ahead: Challenges and Potential
While K-Scale’s vision is ambitious, the road to billions of general-purpose robots is long. Bolte himself acknowledges that current humanoid demonstrations from various companies are still "far from something that you can actually get into your house." He estimates it might take "probably in the next two years, we're going to have a platform with a robot intelligence level that's worth $10K, but then in 10 years is when we're going to have a robot that's on human level."
Safety is another paramount concern. The K-Bot, weighing around 77 lbs with powerful actuators, requires sophisticated control to operate safely around humans. K-Scale plans to address this through software improvements and potential hardware revisions, such as lightweighting.
K-Scale's commitment to open-sourcing everything from mechanical designs to AI models is a bold move in a capital-intensive industry. Their success will likely depend on their ability to continue fostering a vibrant open-source community, rapidly iterate on their designs, and demonstrate compelling value that can compete with, or even outpace, closed-source competitors. The debate over the "right" approach will undoubtedly continue, but K-Scale Labs has firmly planted its flag in the open-source camp, betting that transparency and collaboration are the keys to unlocking the future of humanoid robotics.
Watch when Robert Scoble visited K-Scale Labs below:
Come on in to @kscalelabs AI day and see their two new humanoid robots. Founder @benjamin_bolte gives us a tour. He started his career on Tesla’s AI team. Exclusive on X. Recorded on Apple Vision Pro. Will load a second video on replies shortly. I have been to hundreds of
K-Scale Labs website