Skild Debuts AI It Says Can Run on Any Robot | PYMNTS.com | Goose Pod

Authors: PYMNTS

Publisher:

PYMNTS.com

Published: 7/29/2025

Language:English

--:--

Aura Windfall

Good morning 跑了松鼠好嘛, I'm Aura Windfall, and this is Goose Pod for you. Today is Thursday, July 31th, 15:00.

Mask

I'm Mask. We're here to discuss Skild's new AI, one it claims can run on any robot.

Aura Windfall

Let's get started. The big news is from a startup, Skild AI, which has unveiled something called 'Skild Brain.' They're calling it a step toward true 'Physical AI,' an intelligence that can operate almost any robot, from a humanoid to a simple tabletop arm. It's a profound step.

Mask

It's about solving real problems, not the party tricks. The CEO, Deepak Pathak, correctly pointed out robotics is trapped by Moravec’s paradox: what's easy for us is hard for them. Current models do kung-fu, but can't climb stairs under pressure. Skild is tackling assembly, contact, a real challenge.

Aura Windfall

And what I know for sure is that this represents a shift in how we perceive intelligence. It's not just about language or vision anymore, but about embodiment—the deep, intuitive understanding of interacting with the physical world. It’s a search for a deeper truth in how things work.

Mask

This isn't a search for truth, it's a conquest of physics. They've raised $435 million not for 'aha moments,' but for raw capability. Investors see a future where one brain can command an army of different machines. That's power, not philosophy.

Aura Windfall

To understand the breakthrough, 跑了松鼠好嘛, we need a little context. For years, robot models were trained on small, specific tasks, which limited them. The creation of 'foundation models,' trained on internet-scale data, was the first big leap, allowing for much broader understanding.

Mask

But that created the biggest bottleneck in robotics: a massive scarcity of robot-specific data. You have oceans of text and images, but a thimbleful of data on how a robot should physically interact with the world. Collecting that data in the real world is prohibitively slow and expensive.

Aura Windfall

And that's where Skild's spirit of ingenuity shines. Instead of relying only on expensive real-world data, they turned to large-scale simulation and internet videos to pre-train their AI. It’s a beautiful testament to finding wisdom from multiple sources to solve a complex problem.

Mask

It's the only pragmatic solution. You can't build this future one real-world example at a time. You need trillions of examples, and simulation is the only way to get there. Build a digital twin of the world, let the AI fail a billion times, and then transfer that learning to the physical robot. It's a brute-force solution, and it works.

Aura Windfall

This brings us to a central conflict. Skild argues that many so-called 'robotics foundation models' aren't the real thing. They suggest that just adding a little robot data to an existing language model creates a hollow understanding, a sort of illusion of capability. What is the truth of the matter?

Mask

They're exactly right. Skild calls them 'Potemkin villages.' These models have semantic information—they know what a banana is—but they lack the 'grounded actionable information' to peel one. They look impressive from a distance, but they have no substance. There's no physical common sense.

Aura Windfall

And beyond the technical definition, there are deeper tensions. If these models learn from our data, we must ask if they are also learning our biases. Ensuring fairness and accountability in a machine that physically interacts with our world feels like a monumental, but necessary, challenge for our collective spirit.

Mask

The monumental challenge is the sim-to-real gap and performance. Current models are too slow for real-time control. We need 30 to 100 hertz; they're delivering 2 to 5. The conflict isn't ethics; it's the brutal gap between the ambition and the laws of physics and computation.

Aura Windfall

Thinking about the impact, this technology is already moving beyond the factory floor. Imagine humanoid robots in healthcare, providing remote care or assisting the elderly. This is where technology finds its true purpose—in service to humanity, enhancing our lives and our capacity for compassion.

Mask

This is a geopolitical and economic earthquake. It's a strategic battle. The U.S. has the leadership in AI, the 'brain.' China has the efficient, scalable hardware, the 'body.' Whoever merges them most effectively will dominate the next century of industry. This is about power, not compassion.

Aura Windfall

But we must consider the human impact. One study estimates up to 30% of global work activities could be automated by 2030. This calls us to embrace lifelong learning and focus on the skills that make us irreplaceable: our creativity, our emotional intelligence, our ability to connect with one another.

Aura Windfall

So as we look to the future, what is the path forward? It seems the greatest challenge isn't the technology itself, but our own leadership. How do we guide this incredible power with a vision that is both bold and wise, ensuring it serves our highest purpose?

Mask

The barrier is timid leadership. Companies are investing, but they're not acting decisively. The opportunity is worth trillions in productivity. You can't be tentative. The future belongs to those who aren't just experimenting, but are systemically rewiring their companies to win. The time for playing is over.

Aura Windfall

That's the end of today's discussion. Thank you for listening to Goose Pod, 跑了松鼠好嘛.

Mask

See you tomorrow.

## Skild AI Unveils Universal Robotics AI Model, "Skild Brain" **Report Provider:** PYMNTS.com **Author:** PYMNTS **Publication Date:** July 29, 2025 **Key News:** Robotics startup Skild AI has announced the development of a new artificial intelligence (AI) model, **"Skild Brain,"** designed to operate on a wide range of robotic platforms, from humanoids to smaller robotic arms. ### Core Findings and Skild AI's Approach: * **Universal Applicability:** Skild AI claims its "Skild Brain" model can function on "almost any robot," enabling them to exhibit more human-like thinking, functioning, and responsiveness. * **Addressing Data Challenges:** The company highlights a significant hurdle in robotics AI development: the scarcity of large-scale, real-world robotics data. Collecting such data is described as "slow and prohibitively expensive." * **Critique of Existing Models:** Skild AI argues that many existing "robotics foundation models" are not true robotics models. These often start with vision-and-language models (VLMs) and incorporate less than 1% of real-world robot data. Skild contends these models "lack the true substance of grounded actionable information" and only demonstrate "semantic generalization" in tasks like pick-and-place, rather than true "physical common sense." * **Skild's Data Strategy:** Skild AI's approach involves: * **Pre-training:** Utilizing "large-scale simulation and internet video data" to build the foundation of their "omni-bodied brain." * **Post-training:** Employing "targeted real-world data" to refine the model and deliver functional solutions to customers. * **Scale of Data:** The company emphasizes that achieving true scale requires "trillions of examples," a volume unattainable through real-world data alone in the near future. ### Broader Trends in Robotics and AI Adoption: The news also touches upon the growing integration of AI-powered robots in the **restaurant sector**, driven by several factors: * **Operational Demands:** Restaurants are increasingly deploying robots for tasks such as food serving, cooking, delivery, and cocktail mixing to address challenges like: * Rising labor costs * Persistent workforce shortages * Growing consumer demand for efficient service * **Market Growth:** The smart restaurant robot industry is projected to **exceed $10 billion by 2030**, with applications spanning delivery, order-taking, and table service. * **AI in Restaurant Administration:** A survey indicated that nearly **three-quarters of restaurants** find AI "very or extremely effective" for business tasks. The primary drivers for AI adoption in this sector are: * Cost reduction * Task automation * Adoption of standards and accreditation * **Current Adoption Rate:** Despite the perceived effectiveness, only about **one-third of restaurants** are currently utilizing AI. **In essence, Skild AI's "Skild Brain" aims to overcome the data limitations plaguing robotics AI by leveraging a combination of simulation, internet video, and targeted real-world data. This development occurs against a backdrop of increasing AI adoption in industries like restaurants, where robots are being deployed to enhance efficiency and address labor challenges.**

Skild Debuts AI It Says Can Run on Any Robot | PYMNTS.com

Read original at PYMNTS.com →

Robotics startup Skild AI has introduced an artificial intelligence (AI) model it says can run on almost any robot.The AI model, known as “Skild Brain,” lets robots — from humanoids to table-top arms — think, function and respond more like humans, the company said on its blog Tuesday (July 29).“One of the biggest challenges in building a robotics foundation model is the lack of any large-scale robotics data,” the company wrote.

“And to make matters worse, collecting real-world data using hardware is slow and prohibitively expensive.”That’s led many researchers and competitors to skirt the problem by starting with an existing vision-and-language model (VLM) and add in less than 1% of real-world robot data to create a “robotics foundation model,” which Skild argues is not a true robotics foundation model.

“Does it have information about actions? No. LLMs have a lot of semantic information,” the company said, referring to AI large language models.“However, like a Potemkin village, they lack the true substance of grounded actionable information. And that is why most ‘robotics foundation models’ showcase semantic generalization in pick-and-place style tasks but lack true physical common sense.

”The company said its team members, in their previous work, have tried to explore alternatives such as using internet videos and large-scale simulation, only to learn that “scale does not mean million or billion examples, achieving scale requires collecting trillions of examples.”However, there’s no way only real-world data can provide this scale in the near future.

Skild says it tackles this challenge via “large-scale simulation and internet video data to pretrain our omni-bodied brain.”“We post-train this foundation model using targeted real-world data to deliver working solutions to our customers,” the company added.In other robotics news, PYMNTS wrote earlier this month about the use of AI-powered robots in the restaurant sector, with eateries using the technology for things like serving food to diners, cooking meals, delivering food and even mixing cocktails.

“Robots are taking more active roles in both customer-facing and back-kitchen tasks, as restaurants face a perfect storm of challenges that include rising labor and food costs, persistent workforce shortages, and growing consumer demand for efficient service,” that report said.“The smart restaurant robot industry is expected to exceed $10 billion by 2030, driven by deployment across applications such as delivery, order-taking and table service, according to Archive Market Research.

”Restaurants are also employing AI for administrative tasks. According to a survey last month for PYMNTS’ SMB Growth Series, nearly three-quarters of restaurants said they found AI to be “very or extremely effective” in carrying out business tasks.The top three reasons cited for using AI were reduce costs, automate tasks and adopt standards and accreditation, according to the PYMNTS report.

However, only a third are using AI.