# Skild AI

> Source: https://aiwiki.ai/wiki/skild_ai
> Updated: 2026-10-28
> Fact-checked: 2026-10-27
> Categories: AI Companies, AI Models, Embodied AI, Robotics
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Skild AI." aiwiki.ai, 28 Oct 2026. https://aiwiki.ai/wiki/skild_ai
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Skild AI** is a robotics [artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) company building a general-purpose [foundation model](https://aiwiki.ai/wiki/foundation_models) for physical embodiments, described by the company as a single "robot brain" capable of controlling many different robot bodies across many different tasks. The startup was founded in 2023 (May 2023, according to Contrary Research) by Carnegie Mellon University professors [Deepak Pathak](https://aiwiki.ai/wiki/deepak_pathak) (chief executive officer) and Abhinav Gupta (president), both researchers with long records in [self-supervised learning](https://aiwiki.ai/wiki/self-supervised_learning), [computer vision](https://aiwiki.ai/wiki/computer_vision), and [robot learning](https://aiwiki.ai/wiki/robot_learning).[1][27] Its central product is the **Skild Brain**, a [robot foundation model](https://aiwiki.ai/wiki/robot_foundation_model) designed to be "omni-bodied": the company says it can control quadrupeds, humanoids, tabletop arms, and mobile manipulators "without prior knowledge of its body form."[2][3] In August 2026 it introduced **S1**, which it calls its flagship robotic foundation model, built to learn new tasks from a single video demonstration.[42] At launch, Pathak framed the ambition in absolute terms: "We believe Skild AI represents a step change in how robotics will be scaled, and has the potential to change the entire physical economy."[12]

<figure class="w-56">
<img src="/images/site-2026-10-28-skild_ai-logo-e7e82c133e74-tile.png" alt="Skild AI logo" width="459" height="119" />
<figcaption>Skild AI logo, from the official website <a href="https://skild.ai/">skild.ai</a>. Trademark of its owner.</figcaption>
</figure>

Skild AI emerged from stealth in July 2024 with a $300 million Series A round that valued it at $1.5 billion.[1][8] In January 2026 it announced a $1.4 billion Series C led by [SoftBank Group](https://aiwiki.ai/wiki/softbank_group), with participation from [NVentures](https://aiwiki.ai/wiki/nvidia_nventures) (NVIDIA's venture arm), Macquarie Capital, and Jeff Bezos (through Bezos Expeditions), at a valuation of more than $14 billion; Pathak told Bloomberg that the company had raised more than $2 billion in total.[2][15] In April 2026 it acquired Zebra Technologies' robotics automation business (formerly Fetch Robotics), including the Symmetry Fulfillment orchestration platform.[6][32] In September 2026 the founders wrote that Skild had passed $100 million in annual recurring revenue ten months after its first commercial deployment, with more than 60 paying customers.[43] Skild AI is headquartered in Pittsburgh, with offices in the San Francisco Bay Area and, since February 2026, Bengaluru, India.[5][10]

## What does Skild AI build?

Skild AI belongs to the group of robotics companies that take a software-first or "horizontal" approach: they build the [AI model](https://aiwiki.ai/wiki/model) that controls a robot, but they do not primarily build the robot hardware itself.[27] The company's thesis is that a single model trained across many bodies and tasks will be more robust than narrow controllers built for one platform; in its words, a controller trained for one robot "memorizes, or 'overfits' to" that robot and fails when something unexpected happens.[3]

The company frames its goal as "any robot, any task, one brain."[7] Its technical approach is closer to the playbook used to train [large language model](https://aiwiki.ai/wiki/large_language_model)s than to classical robotics: pre-train on large-scale physics simulation and internet videos of humans, then post-train with targeted real-world data such as [teleoperation](https://aiwiki.ai/wiki/teleoperation) and deployment data.[4][2] Contrary Research groups Skild with other "horizontal" companies such as [Physical Intelligence](https://aiwiki.ai/wiki/physical_intelligence), Covariant, and NVIDIA's [Isaac GR00T](https://aiwiki.ai/wiki/isaac_gr00t) effort, and contrasts it with "vertical" companies such as [Figure AI](https://aiwiki.ai/wiki/figure_ai), [1X Technologies](https://aiwiki.ai/wiki/1x_technologies), and Tesla, which build both the robot and the model.[27]

Pathak has repeatedly invoked [Moravec's paradox](https://aiwiki.ai/wiki/moravecs_paradox) to describe the problem. When the Skild Brain was unveiled in July 2025 he said: "Robotics is marred by Moravec's paradox: the hard problems are easy and the easy problems are hard. A lot of current robotics models focus on tasks that are hard for humans and easy for robots: dancing, kung-fu, because they are free-space actions and do not require any generalization."[9] In a live interview with Alex Heath on the sidelines of the World Economic Forum in Davos in January 2026, he argued that physical intelligence, rather than language, is the real foundation of [artificial general intelligence](https://aiwiki.ai/wiki/artificial_general_intelligence), and that Tesla's self-driving dataset would not help its Optimus robot generalize.[30]

<figure class="embedvideo">
<iframe src="https://www.youtube-nocookie.com/embed/lC1M_Zaje9o" title="Building the general-purpose robotic brain"></iframe>
<figcaption>"Building the general-purpose robotic brain", official video from Skild, published July 29, 2025. <a href="https://www.youtube.com/watch?v=lC1M_Zaje9o">Watch on YouTube</a>.</figcaption>
</figure>

## Who founded Skild AI?

### Co-founders

Deepak Pathak is the chief executive officer. His CMU curriculum vitae lists him as the Raj Reddy Associate Professor in Robotics at [Carnegie Mellon University](https://aiwiki.ai/wiki/cmu)'s Robotics Institute, a post he has held since September 2020.[33] He completed his PhD in computer science at UC Berkeley (2014 to 2019), advised by Alexei A. Efros and Trevor Darrell, and his BTech at IIT Kanpur, where he received the department gold medal in 2014.[33] Between September 2019 and August 2020 he was a researcher at Facebook AI Research with Jitendra Malik and a visiting postdoctoral researcher at Berkeley with Pieter Abbeel.[33] His honors listed in the CV include the Sloan Research Fellowship (2025), the MIT Technology Review TR 35 Under 35 Innovator Award (2024), the Raj Reddy Chair in Robotics at CMU (2024), the Okawa Research Award (2022), and an ONR Young Investigator Award (2026).[33] His best-known papers include "Curiosity-driven Exploration by Self-supervised Prediction" (ICML 2017) and "Learning to Control Self-Assembling Morphologies" (NeurIPS 2019).[35][36]

Abhinav Gupta is the president and co-founder. He came to the CMU Robotics Institute in 2009 as a postdoctoral researcher and joined its faculty in 2011, with an appointment in the Machine Learning Department as well; he holds a bachelor's degree from IIT Kanpur and a PhD in computer science from the University of Maryland.[58] In 2018 CMU announced that he would join a new Facebook AI Research (FAIR) lab in Pittsburgh, led by CMU professor Jessica Hodgins, working part-time with Facebook while keeping his CMU teaching and research duties; his homepage notes that he was back full-time at CMU from May 2022.[34][59] His research has focused on self-supervised visual learning and large-scale visual learning, including the Never Ending Image Learner (NEIL) project, and his awards include a 2016 Sloan Research Fellowship, the PAMI Young Researcher Award, the 2018 ONR Young Investigator Award, the 2020 J.K. Aggarwal Prize, and a Bosch Young Faculty Fellowship.[58][59] At launch Gupta described the social rationale for the company in terms of the labor market: by enabling robots to safely perform automated tasks in any environment, "we can expand the capabilities of robots, democratize their cost, and support the severely understaffed labor market."[12]

According to Contrary Research, Pathak and Gupta had discussed launching a company together for nearly a decade, including while working at FAIR, and decided in early 2023 that the timing was right.[27] Sequoia partner Stephanie Zhan, who first met the founders while they were raising their initial seed round, has described their pitch as one she hears "once in a blue moon."[21] In Skild's Series A announcement, Zhan said: "A GPT-3 moment is coming to the world of robotics," and Sequoia's own essay argued that Skild "has the potential to build in the real world what OpenAI has built in the digital world."[1][21]

### Team and offices

Skild's Series A announcement said the team included robotics and AI experts from Meta, Tesla, [Nvidia](https://aiwiki.ai/wiki/nvidia), Amazon, [Google](https://aiwiki.ai/wiki/google), and universities including Carnegie Mellon, Stanford, UC Berkeley, and UIUC.[1] The Robot Report wrote in July 2025 that the company had grown to more than 25 employees.[9] The company is headquartered in Pittsburgh, with an office in the San Francisco Bay Area and, from February 2026, an office in Bengaluru, India.[10][5] The Bengaluru opening was Skild's first expansion outside the United States; the company said the office would recruit robotics engineers, AI scientists, and systems engineers and work on components of the brain "from perception and control policies to evaluation and deployment infrastructure."[5]

## How much funding has Skild AI raised?

Skild AI is one of the most heavily funded software-focused robotics startups. Pathak told Bloomberg in January 2026 that the company had raised more than $2 billion to date; Crunchbase separately tallied more than $1.83 billion.[15][29] The company's valuation rose from $1.5 billion at the Series A to more than $14 billion at the Series C.[1][2] As of October 2026, no later priced round had been announced.

| Round | Date | Amount | Valuation | Lead investor(s) | Other investors |
|---|---|---|---|---|---|
| Seed | 2023 | Not publicly disclosed | Not disclosed | Not disclosed | Sequoia's Stephanie Zhan has said Sequoia partnered with Skild at the seed round[1] |
| Series A | July 9, 2024 | $300 million | $1.5 billion | [Lightspeed Venture Partners](https://aiwiki.ai/wiki/lightspeed_venture_partners), [Coatue Management](https://aiwiki.ai/wiki/coatue_management), [SoftBank Group](https://aiwiki.ai/wiki/softbank_group), Jeff Bezos (Bezos Expeditions) | Felicis Ventures, [Sequoia Capital](https://aiwiki.ai/wiki/sequoia_capital), [Menlo Ventures](https://aiwiki.ai/wiki/menlo_ventures), [General Catalyst](https://aiwiki.ai/wiki/general_catalyst), CRV, SV Angel, [Carnegie Mellon University](https://aiwiki.ai/wiki/cmu), Amazon Industrial Innovation Fund, Alexa Fund |
| Series B | Mid-2025 (reported dates vary between April and June 2025) | Not officially disclosed; reported as $135 million (Crunchbase) or about $500 million (PitchBook, Technical.ly) | About $4.5 billion (Bloomberg) or $4.7 billion (PitchBook, Technical.ly) | [SoftBank Group](https://aiwiki.ai/wiki/softbank_group) (reported) | Reported: NVIDIA (about $25 million), [Samsung](https://aiwiki.ai/wiki/samsung) (about $10 million), LG Technology Ventures, Hanwha, Mirae Asset, Amazon |
| Series C | January 14, 2026 | $1.4 billion | More than $14 billion | [SoftBank Group](https://aiwiki.ai/wiki/softbank_group) | [NVentures](https://aiwiki.ai/wiki/nvidia_nventures), Macquarie Capital, Bezos Expeditions; Lightspeed, Felicis, Coatue, and Sequoia (follow-on); strategic: [Samsung](https://aiwiki.ai/wiki/samsung), LG, Schneider Electric, CommonSpirit, Salesforce Ventures; others: Disruptive, 1789 Capital, IQT, TF Capital, Andra Capital, Palo Alto Growth Capital, KIC, Alpha Square, Mirae Asset, Destiny |

The Series A was announced on July 9, 2024, when the company came out of stealth; Carnegie Mellon University participated as an investor.[1][12] In late January 2025, Bloomberg and the Financial Times reported that SoftBank was negotiating to lead a $500 million investment at a valuation of about $4 billion.[13][28] Skild never formally announced the resulting Series B, and reports on it differ. In June 2025, SamMobile, citing Bloomberg, reported that the round was led by a $100 million investment from SoftBank at a $4.5 billion valuation, with NVIDIA investing $25 million and Samsung $10 million, and with LG, Hanwha, and Mirae Asset each putting in between $5 million and $10 million.[50] Technical.ly reported in July 2025 that Skild had raised a $500 million Series B in April at a $4.7 billion valuation, while TechCrunch, citing PitchBook data, reported a $500 million round in May at $4.7 billion; Crunchbase recorded it as a $135 million Series B at $4.5 billion.[24][14][29] The Robot Report wrote in July 2025 that Skild had raised a total of $435 million across two rounds.[9]

In December 2025, TechCrunch, citing Reuters, reported that SoftBank and NVIDIA were in talks to lead an investment of more than $1 billion at a $14 billion valuation.[14] The Series C was announced on January 14, 2026: $1.4 billion led by SoftBank at a valuation of more than $14 billion.[2][15] In the same announcement the company said its "live revenue grew from zero to about $30M in just a few months in 2025."[2] Crunchbase News noted that the raise came just over seven months after the Series B, roughly tripling the valuation.[29]

The Series C drew a wide mix of strategic and financial backers. Alongside SoftBank and NVentures, the company named Macquarie Capital, Bezos Expeditions, IQT, Mirae Asset, and Andra Capital among its investors, and The Robot Report listed Samsung, LG, Schneider Electric, CommonSpirit, and Salesforce Ventures as strategic investors.[2][10] Rita Waite, a partner at IQT, said: "Solving intelligence for the physical world unlocks enormous commercial value and long-term strategic national importance."[10]

## What is Skild AI's mission and technical approach?

Skild AI describes its mission as building "intelligence that is grounded in the physical world," with a long-term goal of AGI "rooted in the physical world."[1] The company argues that many so-called robotics foundation models start from a vision-language model and add "less than 1% of real-world robot data," which in its view gives them semantic knowledge but not "true physical common sense."[4]

The technical strategy has several distinguishing pieces:

- **Omni-bodied training.** Skild says it "created a universe with 100,000 different robots" in simulation and trained one model to control them all. Because the model cannot memorize a solution for any single body, the company argues, it has to learn strategies that transfer.[3]
- **Mixed-source data.** The Series C announcement names four data sources: large-scale simulation ("trillions of synthetic experiences"), internet videos of human actions, teleoperation, and real-world deployments.[2] At the Series A, the company said its model was trained on a dataset "1,000 times larger" than those used by rival robotics startups.[12][37]
- **Hierarchical policy.** The Skild Brain pairs a low-frequency high-level manipulation and navigation policy with a high-frequency low-level policy that outputs joint angles and motor torques.[4]
- **In-context adaptation.** In the company's demonstrations, the model adapts within "milliseconds to minutes, depending on the severity of the change," for example when a quadruped's leg is shortened, its knees are locked, or its wheels are jammed.[3]
- **Post-training for deployment.** For new industrial tasks, Skild says the base model is fine-tuned with a small amount of task data in a post-training step; S1, introduced in August 2026, is designed to reduce that step by taking a video demonstration as its prompt.[7][42]

Skild's January 2026 post "Learning by watching human videos" argued that teleoperation alone cannot reach foundation-model scale and described fine-tuning the model to perform new skills by watching videos plus "only a small amount of robot data (<1 hr)."[54] RoboHorizon described a related demonstration in which a robot arm made scrambled eggs after learning the skill from human video.[40]

### What environments will Skild AI deploy in first?

Skild describes a phased path through environments of increasing difficulty. It first deploys in semi-structured settings such as factories and other industrial sites, uses the resulting data to move into less structured places such as hospitals and hotels, and eventually aims at general-purpose robots that work in fully unstructured environments, including homes.[7][11] Pathak and Sequoia's Zhan gave the same staged timeline at Fortune Brainstorm AI, where Pathak argued that the company that deploys first will win by building a "data flywheel."[41]

### The data flywheel

The data flywheel is central to the company's commercial pitch. Because the Skild Brain is omni-bodied, the company says, it "can ingest data from different robots and use it to improve itself," so that robots in the field gather data for the next generation of the model.[7] In March 2026 Pathak said: "If we partner with robotic OEMs (original equipment manufacturers) that already have hundreds of thousands of robots deployed, it gives us a path to extreme scalability and establishes the data flywheel."[52]

In a September 2026 essay, Pathak and Gupta described the strategy in more detail: start from a foundation model trained across robots, tasks, and embodiments; deploy it into specialized applications, using in-context learning to adapt and deployment data to specialize where needed; then bring data from across those deployments back into the general model so that "the next deployment starts from a stronger model."[43] NVIDIA's account of the S1 launch added that experience from Skild's commercial deployments can inform the broader model "where customer agreements permit."[45]

## S1 in-context manipulation model

On August 18, 2026, Skild introduced S1, which it describes as its flagship robotic foundation model, "built from the ground up as an in-context learner." S1 takes a video demonstration as its task prompt. Skild said the demonstrations in its release used one set of model weights without task-specific fine-tuning or post-training. The company showed four tasks that it described as absent from S1's pretraining task set: plant potting, pancake cooking, pour-over coffee making, and kit assembly. Skild said the tasks ran for up to ten minutes; for plant potting, it reported an eleven-minute interval between recording the demonstration and starting autonomous execution.[42]

Skild also reported an internal comparison with a language-conditioned vision-language-action policy. The company said the two policies used the same training data, architecture apart from prompt embedding, and compute across datasets ranging from 1,000 to 100,000 hours. On its internal unseen-task suite at the largest scale, Skild reported average per-step success rates of 66 percent for S1's in-context policy and 9 percent for language prompting. Human intervention recovered failed rollouts so that every step could be graded, which means the figures are not end-to-end autonomous task-completion rates.[42] Skild also estimated that a single in-context demonstration was worth roughly 380 post-training demonstrations for a conventionally post-trained policy, which it said would take 50 to 100 hours of teleoperation to collect.[42][45]

The S1 announcement says Skild carried lessons from LocoFormer into manipulation; LocoFormer is a 2025 locomotion model (Min Liu, Deepak Pathak, and Ananye Agarwal, accepted to CoRL 2025) that adapts to previously unseen legged and wheeled robots by extending its context across episodes.[42][56] Skild said S1 is built on NVIDIA AI infrastructure, and NVIDIA wrote that the model was trained and validated with tools including Isaac Lab, Isaac Sim, Cosmos, and the Newton physics engine, and that the two companies were jointly developing GPU-accelerated contact simulation solvers to be released as part of Newton.[42][45]

As of August 28, 2026, Skild had not released a peer-reviewed S1 paper, model weights, evaluation code, or a public benchmark reproduction. No independent reproduction of the demonstrations or internal study had been published. The results therefore describe Skild's own tasks, protocol, and evaluation rather than established performance across unseen manipulation tasks generally.[42] The S1 post said it was "the first in a series" and that future posts would cover how S1 is trained.[42]

### Physical self-play (September 2026)

On September 23, 2026, Skild published a post-training result it called "physical self-play." Starting from S1, a policy was given a single objective in simulation, to score a soccer goal, and trained by playing against recent versions of itself in NVIDIA Isaac Sim. Skild said behaviors such as getting up after falls, dribbling past a defender, shielding the ball, and tackling emerged without hand-crafted rewards for each, and that after 140 years of simulated play it transferred the policy to a physical humanoid robot.[44] The company argued that "because S1-class models are pre-trained on human data, they are capped at human capability" and that self-play could let robots exceed it, and said it was extending the approach to everyday robotics tasks.[44] Trade coverage in October 2026 reported that early four-agent soccer experiments had begun to produce passing and coordination, according to the company.[47]

## Technology and research foundations

Skild AI's technology builds on research from the Pathak and Gupta labs at CMU and Berkeley, much of it published before the company was founded; the company's July 2025 technical post cites the founders' earlier work on self-supervised robot data collection, learning from internet videos, and large-scale simulation.[4] "Curiosity-driven Exploration by Self-supervised Prediction" (Pathak et al., ICML 2017) introduced an intrinsic curiosity signal that rewards an agent for reaching states whose consequences it cannot yet predict.[35] "Learning to Control Self-Assembling Morphologies" (Pathak et al., NeurIPS 2019) studied how primitive limb agents can link up into composite bodies, using modularity as a route to generalization; its "one policy, many bodies" framing is an early version of the cross-embodiment problem the Skild Brain targets.[36] Gupta's earlier work includes large-scale self-supervised grasping ("Supersizing self-supervision: Learning to grasp from 50k tries and 700 robot hours," ICRA 2016, with Lerrel Pinto) and the NEIL project for learning visual knowledge from web images.[4][59] Pathak's group also published applied work on legged locomotion, including "RMA: Rapid Motor Adaptation for Legged Robots" (2021) and legged locomotion over challenging terrain from egocentric vision.[33][4]

### How is the Skild Brain built?

Publicly disclosed details of the Skild Brain remain limited; the company has not published a formal [model card](https://aiwiki.ai/wiki/model_card) or a full technical report. The company describes a hierarchical architecture with two levels: (1) a low-frequency high-level manipulation and navigation action policy, which provides inputs to (2) a high-frequency low-level action policy that "translates high level commands into precise joint angles and motor torques."[4] NVIDIA's case study describes the same split as a high-level decision-maker that decides what the robot should do and a low-level controller that handles precise movements.[22] An August 2025 Skild post focused on the low-level controller, describing end-to-end humanoid locomotion "driven entirely by online vision and proprioception," with no separate stair or step-over mode.[53]

The model is not told which body it is controlling. In one demonstration the company described, the brain was switched on while a quadruped stood upright and "decide[d] to treat this robot as a small humanoid"; it fell, but succeeded on the third try after the previous trials were prepended to its context as a prompt.[3]

For data, the dominant pre-training sources are simulation and human video. NVIDIA's case study says Skild uses Isaac Lab for reinforcement learning scenarios and Cosmos Transfer to augment training data, which lets the company "acquire a millennium of experience within days."[22] The case study also claims a tenfold reduction in total cost of ownership, citing deployments on robots costing $4,000 to $15,000 compared with conventionally customized systems costing more than $250,000.[22]

## What compute infrastructure does Skild AI use?

In March 2025, Skild announced that after initially using only public cloud services it had moved to a private AI-as-a-service environment from STN, an HPE partner, through STN's GPU One service. The setup is based on HPE Cray XD670 systems with NVIDIA HGX H200 for model training, plus HPE ProLiant Compute DL380a Gen12 servers with eight NVIDIA L40S GPUs to support model visualization.[23][57] STN's co-founder later wrote that the GPU One infrastructure supporting Skild is housed in CoreSite's CH2 [data center](https://aiwiki.ai/wiki/data_center) in Chicago, which he described as home to GPU One's 2.4 MW of data center infrastructure; STN says the service lets startups such as Skild go straight from model training to live deployment.[39]

For simulation, Skild relies heavily on NVIDIA software. The company uses Isaac Lab and Isaac Sim for physics simulation and [NVIDIA Cosmos](https://aiwiki.ai/wiki/nvidia_cosmos) for synthetic [data augmentation](https://aiwiki.ai/wiki/data_augmentation).[7][22] NVIDIA said in September 2026 that Skild also uses NVIDIA Nsight tools for profiling and TensorRT for inference optimization.[45] That tooling overlap, plus NVIDIA's investment in Skild, helps explain why NVIDIA is a partner even though it also develops Isaac GR00T foundation models for humanoids.[27]

## What is the Skild Brain?

The central product is the **Skild Brain**, which the company calls "the industry's first unified robotics foundation model."[2] On its website Skild says it builds robot applications on the brain "to keep us grounded," naming a security and inspection robot platform, a mobile manipulation platform whose low-level skills (grasping, handover, navigation) "are abstracted away using an API call," and an autonomous packing application.[55]

### July 2025 public unveiling

In late July 2025, Skild AI published a technical post and a video showing the Skild Brain for the first time.[4][9] Technical.ly noted that the company had been quietly working on the model since 2023 and had stayed quiet after its most recent raise.[24] The footage showed humanoid, quadruped, and tabletop-arm robots; Technical.ly's screenshots included a humanoid reaching for bagels, robot arms sorting dishes, and a robot sorting ear buds.[24] Pathak said: "Skild AI models can not only solve these easy tasks but also solve everyday hard tasks such as climbing stairs even under adversarial conditions, or assembling fine-grained items, which require vision and reasoning about contact dynamics."[9]

A September 2025 follow-up post described in-context adaptation tests in which the brain controlled robots excluded from its training data. After a robot's calves were removed, it learned to walk after seven to eight seconds; with its knees locked it walked on three legs after two to three seconds of adaptation; and when its wheels were jammed it switched to a walking gait and switched back when they were released.[3]

### Safety design

Skild has said little in public about the safety engineering of the Skild Brain. At the July 2025 unveiling, the company said the model is designed to be safe around humans; Technical.ly, summarizing the release, wrote that it is "trained to apply low force and is highly adaptive to human interactions."[9][24] In the Pittsburgh airport pilot, a Skild team member said the quadruped stops moving if a traveler blocks any of its sensors.[48]

## Manufacturing partnerships and industrial deployments

### NVIDIA, Foxconn, ABB, and Universal Robots (March 2026)

In March 2026 Skild AI announced partnerships with ABB Robotics, Teradyne Robotics' Universal Robots (UR) and Mobile Industrial Robots (MiR), and NVIDIA.[7]

The most concrete announcement involved [Foxconn](https://aiwiki.ai/wiki/foxconn)'s assembly lines in Houston, Texas, where NVIDIA's Blackwell GPU server racks are built.[52] Skild said it would use its brain to control dual robotic arms on these lines. The demonstrated task was to pick and place a busbar, then a limit block, drive 16 screws in succession, and remove the limit block, a task Skild said is "currently done by humans at a line in a Foxconn factory"; the company said the model was fine-tuned with a small amount of robot data and used in-context memory to run a task spanning several minutes.[7] In the accompanying press release Pathak said: "By training an omni-bodied intelligence that transfers skills across embodiments and environments, we're shifting from programming tasks to building systems that continuously learn and improve, even during deployment."[20]

Skild also announced integration partnerships with ABB Robotics and Universal Robots.[7] Marc Segura, president of ABB Robotics, said the integration would "help customers scale industrial-grade automation more quickly and address increasingly complex application scenarios across a broad range of industries." Universal Robots CEO Jean-Pierre Hathout said working with Skild and NVIDIA would bring "advanced AI capabilities to our cobots."[7] NVIDIA's Deepu Talla said Skild's brain "demonstrates how foundation models trained in simulation can be deployed on real robots at scale."[7]

### Which robot platforms does the Skild Brain support?

The brain runs across a broad set of robot bodies in Skild's public demos and deployments. The table below summarizes platforms that the company or its partners have shown or named.

| Platform class | Example bodies | Use cases | Status |
|---|---|---|---|
| Quadrupeds | Legged and wheeled quadrupeds | Security, inspection, airport terminal monitoring | Demonstrated and deployed; Pittsburgh International Airport pilot from September 2026[3][48][55] |
| Humanoids | Humanoids in Skild demos | Locomotion, manipulation, soccer via self-play | Demonstrated[44][53] |
| Industrial arms | ABB robots | Industrial automation | Partner integration announced March 2026[7] |
| Collaborative arms | Universal Robots cobots | Variable cobot tasks | Partner integration announced March 2026[7] |
| Dual-arm manipulators | Foxconn assembly cells | Blackwell server assembly (busbar, screws) | Deployment announced March 2026[7][43] |
| [Autonomous mobile robot](https://aiwiki.ai/wiki/autonomous_mobile_robot)s | MiR robots; Fetch robots from the Zebra acquisition | Material movement, warehouse logistics | MiR partnership; Fetch line acquired April 2026[7][11] |

## Who are Skild AI's customers?

The Series C announcement said Skild's robots were deployed in security, construction, delivery, data centers, warehouses, and factory assembly, and that live revenue had grown from zero to about $30 million in a few months in 2025.[2] By September 2026, Pathak and Gupta wrote that the company had "scaled to 60+ paying customers" in ten months, had crossed $100 million in annual recurring revenue, and had recognized $50 million in revenue since starting deployments. They said about 90 percent of revenue came mainly from manipulation tasks and 10 percent from mobility, including 4 percent from the Zebra/Fetch AMR business.[43][46]

Named deployments and pilots include:

- **Foxconn and NVIDIA:** the Skild Brain on dual-arm manipulators for high-precision assembly of NVIDIA Blackwell systems.[43][45]
- **Sumitomo Wiring Systems:** Skild said it was "working towards deploying S1" to automate wire harness manufacturing processes considered hard to automate.[43]
- **Mitsui & Co.:** a pilot of S1-powered general-purpose robots in commercial kitchens; Skild said Mitsui's supply chains serve 1.4 million meals a day across Japan.[43]
- **Pittsburgh International Airport:** a pilot announced September 29, 2026 through the airport's xBridge innovation program, with up to four Skild-powered quadrupeds and two charging stations in the Airside Terminal. Technical.ly described it as a six-month pilot. The first phase covers mobile indoor air-quality monitoring; later phases are to explore detecting full trash bins, spills, and passenger congestion, and possibly pairing the robot with BioFlyte's airborne biological threat detection and autonomous food and beverage delivery. Skild handles installation, mapping, supervision, and monitoring and trains airport staff.[48][49]

In 2025 TechCrunch also reported strategic partnerships with LG CNS and Hewlett Packard Enterprise.[14]

### Zebra Technologies acquisition (April 2026)

On April 15, 2026, Skild AI acquired Zebra Technologies' Robotics Automation business, including the Symmetry Fulfillment orchestration platform, which coordinates tasks between robots and frontline workers using Zebra wearable devices.[6][32] The financial terms were not disclosed; Zebra said that in addition to cash consideration it received an equity stake in Skild AI.[32][11]

The acquired business traces back to Fetch Robotics, an autonomous mobile robot company founded in 2014 by Melonee Wise. Zebra acquired the remaining 95 percent of Fetch, of which it already owned 5 percent, for $291 million in July 2021.[11] On December 9, 2025, Zebra determined that it would "dispose or exit" its robotics automation solutions business, and said it expected up to $80 million of one-time pre-tax charges, including about $60 million of non-cash asset impairment charges.[51]

Pathak told The Robot Report that "the Fetch Team is the main reason for the acquisition as they bring years of deployment experience." Skild plans to support the existing Fetch/Zebra installed base, sell new Fetch robots, and integrate the Skild Brain into Fetch products.[11] Skild and Zebra described the goal as an end-to-end [warehouse automation](https://aiwiki.ai/wiki/warehouse_automation) offering: humanoids for pick-and-place, robot dogs for inspection, robotic arms for packing, AMRs for material movement, and an orchestration layer to control them all.[38]

## How does Skild AI compare with its peers?

Contrary Research's 2024 profile divides Skild's competitors into horizontal companies, which build models or platforms for other companies' robots, and vertical companies, which build both the robot and its AI.[27]

| Company | Approach (per Contrary Research) | Notes |
|---|---|---|
| Skild AI | Horizontal: omni-bodied foundation model (Skild Brain) | Does not primarily build robot hardware |
| [Physical Intelligence](https://aiwiki.ai/wiki/physical_intelligence) | Horizontal: foundation models and learning algorithms for general-purpose robots | Founded 2024 |
| Covariant | Horizontal: Covariant Brain for warehouse picking and sorting | Amazon hired its founders and licensed its models in August 2024 |
| NVIDIA [Isaac GR00T](https://aiwiki.ai/wiki/isaac_gr00t) | Horizontal: foundation models and tools for humanoids | Also a Skild investor and partner |
| [Sanctuary AI](https://aiwiki.ai/wiki/sanctuary_ai) | Humanoid (Phoenix) plus AI control system (Carbon) | Listed by Contrary among horizontal competitors |
| [Figure AI](https://aiwiki.ai/wiki/figure_ai) | Vertical: humanoid robots with in-house AI | Hardware plus model |
| [1X Technologies](https://aiwiki.ai/wiki/1x_technologies) | Vertical: humanoids NEO and EVE with in-house AI | Hardware plus model |
| [Tesla Optimus](https://aiwiki.ai/wiki/tesla_optimus) | Vertical: general-purpose humanoid | Hardware plus model |

Contrary notes that Amazon hired Covariant's founders and about a quarter of its employees in August 2024 and signed a non-exclusive license to its robotic foundation models.[27] The Robot Report has also listed Qualcomm and X Square Robot among companies working on general-purpose AI for robotics.[10]

The March 2026 agreements with ABB Robotics and Universal Robots made two large robot makers distribution partners for the Skild Brain rather than only potential competitors.[7]

## Strategic positioning

Skild's go-to-market is enterprise-led. At the Series A, SiliconANGLE reported that the founders intended a business model similar to OpenAI's: selling access to the robot brain and letting customers fine-tune the foundation model for their own tasks.[12] By 2026 the company was also selling complete applications: the Zebra acquisition gave Skild a deployed fleet management platform and a customer base in fulfillment, and the founders' September 2026 essay said Skild had "focused on deployment as a core part of the technology itself and not the outcome of it."[6][43]

The same essay argued that demonstration videos are a poor measure of progress, because "a successful clip from a robot with 5%, 10%, or 99% accuracy can look exactly the same," and that speed matters as much as accuracy on a production line: "A robot that is 99.9% accurate but ten times too slow is not almost deployable. It's not deployable."[43]

### Labour market thesis

The company has consistently framed its market in terms of labor shortages. Its Series A announcement cited projections of 2.1 million unfilled US manufacturing jobs by 2030.[1] At Davos in January 2026, Pathak told Business Today that "in the near future, we see enterprise-like jobs getting filled by robots, more and more than we see today," pointed to "over a million unfilled jobs globally" in work "people do not want to do," and said India was "a strong contender" to become a manufacturing hub beyond China.[31]

## Industry context

Skild AI is part of a 2023 to 2026 wave of robotics startups built on the idea that the recipe behind large language models will work for robots. The wave followed the appearance of [vision-language-action model](https://aiwiki.ai/wiki/vision_language_action_model)s such as RT-2 and PaLM-E (see [PaLM-E](https://aiwiki.ai/wiki/palm-e_an_embodied_multimodal_language_model)), NVIDIA's push into simulation and its Isaac GR00T platform for humanoids, and renewed investor interest in embodied AI. Crunchbase News reported that robotics startups raised $13.8 billion in 2025, up from $7.8 billion in 2024 and above the $13.1 billion raised in 2021.[29] Other heavily funded robot-brain companies in the period included Physical Intelligence, which TechCrunch said had reportedly raised $600 million at a $5.6 billion valuation in late 2025.[14]

## What are the risks and limitations?

Generalization across robot bodies is hard, and Skild's own executives have said demos overstate progress: in their September 2026 essay the founders wrote that it took a week to cook the first egg with their model "then two months to make it work reliably with different eggs and in different setups," and that "the amount of effort to squeeze the last 5% of performance greatly exceeds that of the first 95%."[43] Real-world safety matters more for [embodied AI](https://aiwiki.ai/wiki/embodied_ai) than for software; a [chatbot](https://aiwiki.ai/wiki/chatbot) error is a nuisance, while a robot that grabs the wrong object near a person can hurt someone.

Competition is intense. Physical Intelligence is building competing general-purpose robot foundation models (its models include [pi0](https://aiwiki.ai/wiki/pi0)), NVIDIA supplies much of the simulation infrastructure Skild uses while developing its own Isaac GR00T foundation models, and vertically integrated humanoid makers such as Figure, 1X, and Tesla build their own AI.[27]

Valuation is far ahead of revenue: the company was valued at more than $14 billion in January 2026, when it reported reaching about $30 million in live revenue, and its annual recurring revenue passed $100 million in September 2026.[2][43] The valuation therefore rests mainly on expected future growth, which is a risk if [scaling laws](https://aiwiki.ai/wiki/scaling_laws) for robotics turn out slower or messier than they did for text.

Skild's financing has also been opaque. In July 2025 Technical.ly reported that the company's recent raises were not backed up by SEC Form D filings under the Skild name: the only "Skild" Form D it found was for a special purpose vehicle, filed in May 2025, logging just over $1 million raised, compared with an $814 million total reported by PitchBook.[24] Published figures for the Series B still differ by source (see the funding section above).

### Open technical questions

Skild has not published a peer-reviewed paper describing the Skild Brain architecture, the training mixture, or its evaluation methodology in detail, which makes it hard for outside researchers to assess claims such as "a millennium of experience within days" or S1's internal benchmark results.[22][42] Skild's S1 post itself argues that evaluating in-context learning requires separating seen from unseen tasks and short from long horizons, and that results depend on how far the test is from training.[42]

A second open question concerns the limits of simulation. Skild says its pre-training relies mainly on simulation and human video and that real-world data is used for post-training, and NVIDIA wrote in 2026 that Isaac Lab and the Newton physics engine help Skild "reduce the simulation-to-reality gap."[4][45] How much real-world data is needed to close that gap for contact-rich manipulation remains one of the central bets the company is making.

## Timeline of key events

| Date | Event |
|---|---|
| 2023 | Company founded by [Deepak Pathak](https://aiwiki.ai/wiki/deepak_pathak) and Abhinav Gupta (May 2023 per Contrary Research) |
| July 9, 2024 | Series A of $300 million announced at a $1.5 billion valuation; Skild emerges from stealth |
| January 2025 | Bloomberg and the Financial Times report SoftBank in talks to lead a $500 million round |
| March 13, 2025 | HPE and STN announce Skild's private AI-as-a-service compute on HPE Cray XD670 with NVIDIA H200 |
| Mid-2025 | Series B led by SoftBank at a reported $4.5 billion to $4.7 billion valuation; amount reported variously as $135 million or $500 million |
| July 29, 2025 | Skild Brain publicly unveiled in a technical post and video |
| September 24, 2025 | "The case for an omni-bodied robot brain" post on training across 100,000 simulated robots |
| December 2025 | Reuters reports SoftBank and NVIDIA in talks to invest at a $14 billion valuation |
| January 14, 2026 | Series C of $1.4 billion at a valuation of more than $14 billion, led by SoftBank |
| January 2026 | Pathak speaks on the sidelines of the World Economic Forum in Davos |
| February 19, 2026 | Bengaluru office opened, Skild's first outside the US |
| March 16 to 19, 2026 | Partnerships with ABB Robotics, Universal Robots, and NVIDIA announced; Foxconn Blackwell assembly deployment |
| April 15, 2026 | Acquisition of Zebra Technologies' Robotics Automation business (formerly Fetch Robotics) and the Symmetry Fulfillment platform |
| August 18, 2026 | S1 in-context manipulation model introduced |
| September 10, 2026 | Founders say Skild passed $100 million in annual recurring revenue with 60+ paying customers |
| September 23, 2026 | "Physical self-play" result: S1-based policy learns soccer in simulation and transfers to a humanoid |
| September 29, 2026 | Six-month quadruped pilot begins at Pittsburgh International Airport |

## Founders' broader contribution to robotics AI

Much of Skild AI's pitch rests on the founders' research careers. Pathak's curiosity-driven exploration paper is a well-known contribution to the literature on intrinsic motivation in reinforcement learning.[35] His work on self-assembling morphologies set up the cross-embodiment framing that the Skild Brain is now scaling, and his later applied papers on legged locomotion, such as rapid motor adaptation for quadrupeds, showed learning-based controllers coping with real-world [distribution shift](https://aiwiki.ai/wiki/distribution_shift)s.[36][33] Gupta's work on self-supervised visual learning, large-scale grasping, and NEIL did similar work on the perception side, and he worked part-time at FAIR's Pittsburgh lab from 2018 until returning full-time to CMU in May 2022.[34][59] Skild AI is betting that methods that worked in their labs at academic scale will keep working at industrial scale with enough simulation, real-world data, and compute. Whether that is right is still an open question.

## See also

- [List of neolabs](https://aiwiki.ai/wiki/list_of_neolabs)
- [Neolab](https://aiwiki.ai/wiki/neolab)
- [Physical Intelligence](https://aiwiki.ai/wiki/physical_intelligence)
- [pi0](https://aiwiki.ai/wiki/pi0)
- [Isaac GR00T](https://aiwiki.ai/wiki/isaac_gr00t)
- [Embodied AI](https://aiwiki.ai/wiki/embodied_ai)
- [Carnegie Mellon University](https://aiwiki.ai/wiki/cmu)
- [Robot learning](https://aiwiki.ai/wiki/robot_learning)
- [Robot foundation models](https://aiwiki.ai/wiki/robot_foundation_model)
- [Foundation models](https://aiwiki.ai/wiki/foundation_model)
- [In-context learning](https://aiwiki.ai/wiki/in_context_learning)
- [Physical AI](https://aiwiki.ai/wiki/physical_ai)

## References

[1] Skild AI. "Announcing our $300M Series A Funding." July 9, 2024. https://www.skild.ai/blogs/announcing-our-300m-series-a
[2] Skild AI. "Announcing Series C." January 14, 2026. https://www.skild.ai/blogs/series-c
[3] Skild AI. "The case for an omni-bodied robot brain." September 24, 2025. https://www.skild.ai/blogs/omni-bodied
[4] Skild AI. "Building the general-purpose robotic brain." July 29, 2025. https://www.skild.ai/blogs/building-the-general-purpose-robotic-brain
[5] Skild AI. "Skild AI Expands Global Footprint To Bengaluru." February 19, 2026. https://www.skild.ai/blogs/bengaluru
[6] Skild AI. "Skild AI Acquires Zebra Technologies' Robotics Arm to Bring Omni-Bodied Intelligence to Warehouses." April 15, 2026. https://www.skild.ai/blogs/skild-zebra
[7] Skild AI. "The Reindustrial Revolution: Partnering with ABB Robotics, Universal Robots, and NVIDIA." March 19, 2026. https://www.skild.ai/blogs/reindustrial-revolution
[8] The Robot Report. "Skild AI grabs $300M to build foundation model for robotics." July 2024. https://www.therobotreport.com/skild-ai-grabs-300m-to-build-foundation-ai-model-for-robotics/
[9] The Robot Report (Mike Oitzman). "Skild AI is giving robots a brain." July 30, 2025. https://www.therobotreport.com/skild-ai-is-giving-robots-a-brain/
[10] The Robot Report. "Skild AI raises $1.4B to build 'omni-bodied' robot brain." January 2026. https://www.therobotreport.com/skild-ai-raises-1-4b-building-omni-bodied-robot-skild-brain/
[11] The Robot Report (Mike Oitzman). "Skild acquires Fetch Robotics assets from Zebra." April 2026. https://www.therobotreport.com/skild-acquires-fetch-robotics-assets-from-zebra-automation/
[12] SiliconANGLE. "Skild AI raises $300M to build a general-purpose, AI-powered brain for any robot." July 9, 2024. https://siliconangle.com/2024/07/09/skild-ai-raises-300m-build-general-purpose-ai-powered-brain-robot/
[13] TechCrunch. "SoftBank to invest $500M in robotics startup Skild AI." January 28, 2025. https://techcrunch.com/2025/01/28/softbank-to-invest-500m-in-robotics-startup-skildai/
[14] TechCrunch. "SoftBank and Nvidia reportedly in talks to fund Skild AI at $14B, nearly tripling its value." December 8, 2025. https://techcrunch.com/2025/12/08/softbank-and-nvidia-reportedly-in-talks-to-fund-skildai-at-14b-nearly-tripling-its-value/
[15] TechCrunch. "Robotics software maker Skild AI hits $14B valuation." January 14, 2026. https://techcrunch.com/2026/01/14/robotic-software-maker-skild-ai-hits-14b-valuation/
[16] Bloomberg. "Robotics Startup Skild Valued Above $14 Billion After SoftBank-Led Funding Round." January 14, 2026. https://www.bloomberg.com/news/articles/2026-01-14/robotics-startup-skild-valued-above-14-billion-after-softbank-led-funding-round
[17] Bloomberg. "Skild AI Acquires Zebra Technologies' Robotics Automation Business." April 15, 2026. https://www.bloomberg.com/news/articles/2026-04-15/skild-ai-acquires-zebra-technologies-robotics-automation-business
[18] Business Wire. "Skild AI Raises $1.4B, Now Valued Over $14B." January 14, 2026. https://www.businesswire.com/news/home/20260114335623/en/Skild-AI-Raises-$1.4B-Now-Valued-Over-$14B
[19] Business Wire. "Skild AI Acquires Zebra Technologies' Robotics Automation Business." April 15, 2026. https://www.businesswire.com/news/home/20260415518240/en/Skild-AI-Acquires-Zebra-Technologies-Robotics-Automation-Business
[20] GlobeNewswire. "Skild AI Expands Generalized Robot Intelligence Across Industries With ABB Robotics, Universal Robots, and NVIDIA." March 17, 2026. https://www.globenewswire.com/news-release/2026/03/17/3256839/0/en/Skild-AI-Expands-Generalized-Robot-Intelligence-Across-Industries-With-ABB-Robotics-Universal-Robots-and-NVIDIA.html
[21] Sequoia Capital (Stephanie Zhan). "Partnering with Skild: The Future of Embodied Intelligence." July 2024. https://sequoiacap.com/article/partnering-with-skild/
[22] NVIDIA. "Skild AI Builds Omni-Bodied Robot Brain With NVIDIA." 2025. https://www.nvidia.com/en-us/case-studies/skild-ai/
[23] HPE. "Skild AI accelerates development of human-like robot brain with AI solutions from Hewlett Packard Enterprise." March 13, 2025. https://www.hpe.com/us/en/newsroom/press-release/2025/03/skild-ai-accelerates-development-of-human-like-robot-brain-with-ai-solutions-from-hewlett-packard-enterprise.html
[24] Technical.ly (Alice Crow and Katie Malone). "Skild AI finally reveals its futuristic robot 'brain', but SEC filings still don't add up." July 29, 2025. https://technical.ly/entrepreneurship/skild-brain-unveiled-ai-human-robot/
[25] Technical.ly. "Skild faces real-world test of its robot brain in Nvidia, Foxconn factory deal." March 2026. https://technical.ly/entrepreneurship/pittsburgh-skild-ai-nvidia-foxconn-robotics-deployment/
[26] Technical.ly. "Robotics startup Skild AI raises $1.4B led by SoftBank Group." January 2026. https://technical.ly/entrepreneurship/skild-ai-1-4-billion-raise/
[27] Contrary Research. "Skild AI: Business Breakdown & Founding Story." 2024. https://research.contrary.com/company/skild-ai
[28] Pittsburgh Post-Gazette. "SoftBank in talks with Pittsburgh robotics startup Skild AI to lead $500 million round." January 29, 2025. https://www.post-gazette.com/business/tech-news/2025/01/29/softbank-pittsburgh-robotics-skild-ai/stories/202501290077
[29] Crunchbase News. "Robotics Startup Skild AI Lands $1.4B, Tripling Valuation To $14B In Just 7 Months." January 2026. https://news.crunchbase.com/venture/robotics-startup-skild-ai-triples-valuation/
[30] Sources / Alex Heath. "Building one AI brain for every robot." February 26, 2026. https://sources.news/p/skild-ai-ceo-robotics-brain-davos
[31] BusinessToday. "Desk jobs will be replaced by robots, says Skild AI CEO Deepak Pathak." January 21, 2026. https://www.businesstoday.in/wef-2026/story/desk-jobs-will-be-replaced-by-robots-says-skild-ai-ceo-deepak-pathak-512060-2026-01-21
[32] Zebra Technologies. "Skild AI Acquires Zebra Technologies' Robotics Automation Business." April 15, 2026. https://www.zebra.com/us/en/about-zebra/newsroom/press-releases/2026/skild-ai-acquires-zebra-technologies--robotics-automation-busine.html
[33] Carnegie Mellon University. "Deepak Pathak Curriculum Vitae." CMU School of Computer Science. https://www.cs.cmu.edu/~dpathak/CV_Pathak.pdf
[34] Carnegie Mellon University. "Hodgins, Gupta Join Facebook AI Research." July 2018. https://www.cmu.edu/news/stories/archives/2018/july/facebook-ai-research.html
[35] Pathak, Deepak; Agrawal, Pulkit; Efros, Alexei A.; Darrell, Trevor. "Curiosity-driven Exploration by Self-supervised Prediction." ICML 2017. https://arxiv.org/abs/1705.05363
[36] Pathak, Deepak; Lu, Chris; Darrell, Trevor; Isola, Phillip; Efros, Alexei A. "Learning to Control Self-Assembling Morphologies: A Study of Generalization via Modularity." NeurIPS 2019. https://arxiv.org/abs/1902.05546
[37] Pittsburgh Startup News. "Skild AI Launches Foundation Model That Could Redefine Robotics." 2025. https://pittsburghstartupnews.substack.com/p/skild-ai-launches-foundation-model
[38] DC Velocity. "Zebra sells off its Fetch AMR division." April 2026. https://www.dcvelocity.com/material-handling/robotics/zebra-sells-off-its-fetch-amr-division
[39] CoreSite. "Watts, Water and Workloads: How STN and CoreSite Built the Infrastructure Behind Skild AI at CH2." https://www.coresite.com/blog/watts-water-and-workloads-how-stn-and-coresite-built-the-infrastructure-behind-skild-ai-at-ch2
[40] RoboHorizon. "Skild AI Teaches Robots to Cook by Making Them Watch YouTube." January 2026. https://robohorizon.com/en-us/magazine/2026/01/skild-ai-teaches-robots-to-cook-by-making-them-watch-youtube/
[41] Fortune. "Robots are really advancing because they're learning to think for themselves and they're close to figuring out door handles, execs say." January 6, 2026. https://fortune.com/2026/01/06/robots-advancing-from-backflips-to-opening-door-handles-skild-sequoia/
[42] Skild AI. "Introducing S1: In-Context Learning for Robotics." August 18, 2026. https://www.skild.ai/blogs/s1
[43] Deepak Pathak and Abhinav Gupta (Skild AI). "The Hidden Pillar of Robotics." September 10, 2026. https://www.skild.ai/blogs/skild-crosses-100m-arr
[44] Skild AI. "Physical Self-Play." September 23, 2026. https://www.skild.ai/blogs/physical-self-play
[45] NVIDIA Blog. "Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video." September 10, 2026. https://blogs.nvidia.com/blog/skild-ai-s1-physical-ai/
[46] Robotics & Automation News. "Skild AI reaches $100 million in annual recurring revenue after 10 months." September 18, 2026. https://roboticsandautomationnews.com/2026/09/18/skild-ai-reaches-100-million-in-annual-recurring-revenue-after-10-months/104958/
[47] Robotics & Automation News. "Skild AI trains robots to learn physical skills through self-play." October 16, 2026. https://roboticsandautomationnews.com/2026/10/16/skild-ai-trains-robots-to-learn-physical-skills-through-self-play/105994/
[48] Technical.ly (Alice Crow). "Robot dogs will soon patrol the Pittsburgh airport to spot spills and monitor air quality." September 29, 2026. https://technical.ly/entrepreneurship/pittsburgh-airport-skild-ai-robot-dogs-pilot/
[49] International Airport Review (Gabriel Higgins). "Picture of the week: Pittsburgh International Airport to test Skild AI quadruped robots for air quality monitoring and terminal operations through its xBridge innovation programme." October 9, 2026. https://www.internationalairportreview.com/picture-of-the-week-pittsburgh-international-airport-to-trial-four-legged-skild-ai-robots-in-airside-terminal-under-xbridge-programme/2136709.article
[50] SamMobile. "Samsung joins NVIDIA in making an investment in a robotics company." June 12, 2025. https://www.sammobile.com/news/samsung-joins-nvidia-in-making-an-investment-in-a-robotics-company/
[51] Zebra Technologies Corporation. Form 8-K (Items 2.05 and 2.06), filed with the US SEC. December 9, 2025. https://www.sec.gov/Archives/edgar/data/877212/000162828025056882/zbra-20251209.htm
[52] CXO Digital Pulse. "Skild AI and Nvidia Deploy 'Robot Brain' on Blackwell Assembly Lines." March 17, 2026. https://www.cxodigitalpulse.com/skild-ai-and-nvidia-deploy-robot-brain-on-blackwell-assembly-lines/
[53] Skild AI. "One Model, Any Scenario: End-to-end Locomotion from Vision." August 6, 2025. https://www.skild.ai/blogs/one-policy-all-scenarios
[54] Skild AI. "Learning by watching human videos." January 12, 2026. https://www.skild.ai/blogs/learning-by-watching
[55] Skild AI. Company website (homepage). Accessed October 27, 2026. https://www.skild.ai/
[56] Liu, Min; Pathak, Deepak; Agarwal, Ananye. "LocoFormer: Generalist Locomotion via Long-context Adaptation." arXiv:2509.23745, 2025 (CoRL 2025). https://arxiv.org/abs/2509.23745
[57] Silicon UK (Business Wire). "Skild AI Accelerates Development of Human-like Robot Brain with AI Solutions from Hewlett Packard Enterprise." March 13, 2025. https://www.silicon.co.uk/press-release/skild-ai-accelerates-development-of-human-like-robot-brain-with-ai-solutions-from-hewlett-packard-enterprise
[58] Carnegie Mellon University Robotics Institute. "Abhinav Gupta Wins Sloan Research Fellowship." February 23, 2016. https://www.ri.cmu.edu/abhinav-gupta-wins-sloan-research-fellowship/
[59] Abhinav Gupta. Personal homepage, Carnegie Mellon University. https://www.cs.cmu.edu/~abhinavg/

