nooma labs
Physics foundation action models for robotics.
Website: https://www.noomalabs.com/
Cover Block
PUBLIC
| Field | Value |
|---|---|
| Name | nooma labs |
| Tagline | Physics foundation action models for robotics. |
| Headquarters | Kyoto |
| Founded | 2026 |
| Industry | Deeptech |
| Technology | Robotics |
Links
Publicly reported
- Website: https://www.noomalabs.com/
Summary and Signal
PUBLIC nooma labs is a newly surfaced Kyoto-based robotics research effort focused on "physics foundation action models" for robotics, and it merits attention because the category has attracted substantial capital and technical interest while the company itself remains only lightly documented in public sources [noomalabs.com, retrieved 2024] [The Robot Report, retrieved 2026]. The public record is thin, but the strongest available thread is that Nooma Labs presents itself as an independent laboratory in Kyoto and, according to Cross Labs archival material, continues the research direction previously led by Olaf Witkowski at Cross Labs [NOOMA Labs, The Space Between Minds, retrieved 2026] [About Cross Labs | Historical Research Archive, retrieved 2026] [A New Chapter: Nooma Labs | Cross Labs Archive, 2026-08-31].
The product claim is straightforward: the company says it builds physics foundation action models for robotics, a framing that places it near the emerging intersection of robot foundation models, embodied learning, and physics-informed control rather than conventional application-layer robotics software [noomalabs.com, retrieved 2024]. That positioning is directionally interesting because well-funded peers in adjacent robot foundation model work, including Physical Intelligence and Field AI, have helped validate investor appetite for platforms that aim to generalize robot behavior across tasks and environments [The Robot Report, retrieved 2026].
The main constraint for investors is evidentiary rather than conceptual. No founders are independently confirmed in the structured public record for noomalabs.com, no funding rounds are confirmed, no customers or deployments are named, and no hiring footprint was surfaced from major public channels in the research provided [noomalabs.com, retrieved 2024].
On team background, the only supported linkage is to Olaf Witkowski's continuing research program as described by Cross Labs archival sources, plus Nooma Labs' own description of its work on agency, cognition, adaptation, collective intelligence, and computation [About Cross Labs | Historical Research Archive, retrieved 2026] [A New Chapter: Nooma Labs | Cross Labs Archive, 2026-08-31] [NOOMA Labs, The Space Between Minds, retrieved 2026]. That is enough to suggest a research-led origin, but not enough to establish commercial readiness, go-to-market shape, or the composition of the operating team.
For the next 12 to 18 months, the practical watchpoints are basic but consequential: whether the company clarifies who is building it, whether it publishes technical results or demonstrations tied specifically to robotics performance, and whether any financing, partnerships, or commercialization model become visible through independent sources. Until then, the opportunity reads as an early research platform with credible thematic alignment to a fast-forming robotics stack, but with company-specific proof points still largely confined to first-party and affiliated archival sources.
No independent source found -- This section relies primarily on company and affiliated archival sources, with independent reporting used for category context rather than direct verification of company operations.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Industry / Vertical | Deeptech |
| Technology Type | Robotics |
Company Overview
PUBLIC
What is verifiable today is narrow but still useful: nooma labs presents itself as a Kyoto-based independent research laboratory working on robotics-oriented foundation models, and the public record around the domain remains thin outside company-controlled materials and a related research archive [noomalabs.com, retrieved 2024] [noomalabs.org, retrieved 2024] [crosslabs.org, retrieved 2026]. The company website states that nooma labs "builds physics foundation action models for robotics," while the separate noomalabs.org site describes Nooma Labs as an independent laboratory in Kyoto and frames its work around agency, cognition, adaptation, learning, collective intelligence, and computation [noomalabs.com, retrieved 2024] [noomalabs.org, retrieved 2024].
The chronology that can be assembled from public sources is modest. [crosslabs.org, retrieved 2026]. That does not establish a legal entity name, financing history, or commercial launch timeline, but it does suggest a research-continuity story rather than a standing-start formation [crosslabs.org, retrieved 2026] [olafwitkowski.com, retrieved 2026].
Several basic company descriptors remain unconfirmed in public filings or databases provided for this section. No founder list, legal entity details, stage, or funding round could be corroborated from the allowed source set, so the safest reading is that nooma labs is an early, lightly surfaced robotics research effort with a clearer scientific identity than operating footprint in the public record [noomalabs.com, retrieved 2024] [noomalabs.org, retrieved 2024].
No independent source found -- This section relies primarily on company-controlled websites and a related research archive, with no corroborating Crunchbase or state filing evidence provided.
The Product and the Stack
Product Surface
MIXED The public record here is unusually thin, so the safest starting point is the homepage itself. The company describes nooma labs as building "physics foundation action models for robotics," and the site pairs that line with the tagline "A world in the making" [noomalabs.com]. On the evidence available, that supports a narrow product description: nooma labs appears to be working on model-based robotics software rather than a deployed hardware system, customer-facing application suite, or publicly documented services layer [noomalabs.com].
The adjacent Kyoto research entity creates some naming ambiguity that matters for product analysis. The noomalabs.org site describes NOOMA Labs as an independent research laboratory in Kyoto and says it continues scientific work associated with Cross Labs, while Cross Labs archive pages state that Olaf Witkowski's research program continues at Nooma Labs and that the laboratory is independent from Cross Compass Ltd. [noomalabs.org] [crosslabs.org, August 2026] [crosslabs.org, retrieved 2026]. Those sources help explain the research framing around agency, cognition, adaptation, collective intelligence, and computation, but they do not independently verify that the noomalabs.com domain offers the same organization, a commercial product, or a production-ready robotics stack [noomalabs.org] [crosslabs.org, August 2026].
Technology Characterization
MIXED The phrase "physics foundation action models" suggests an attempt to combine physical world modeling with action generation for robots, a direction that sits inside the broader push toward robotics foundation models documented elsewhere in the sector [noomalabs.com] [The Robot Report, retrieved 2026]. That said, the available public materials do not specify model architecture, training data, robot embodiments, simulation environment, deployment method, evaluation benchmarks, or customer use cases, so any deeper technical characterization would move beyond what the sources support [noomalabs.com].
Just as important, there is no verified public demo, technical paper, benchmark sheet, or product documentation in the supplied materials for the noomalabs.com entity. The result is that the company's core idea is legible, but the implementation remains largely unobservable from public sources. For readers, the central diligence question is not whether robotics foundation models are a real category, they are, but whether this specific team has translated that framing into a differentiated system that can be tested outside homepage language [The Robot Report, retrieved 2026] [noomalabs.com].
No independent source found -- This section relies primarily on company-owned websites and archival pages, with no independent product demo, benchmark, funding disclosure, or customer evidence in the public record used here.
The Market They Are Entering
PUBLIC
The market matters now because robotics investors and operators have shifted from isolated task automation toward foundation-model approaches that aim to generalize across hardware, environments, and workflows, but the commercial contours are still forming [The Robot Report, retrieved 2026].
Public evidence specific to nooma labs is thin, so the market lens has to start one layer out, with analogous public reporting on robotics foundation models rather than company-specific demand data. The clearest signal is capital formation around what The Robot Report describes as robot foundation models and "physical AI" infrastructure, including Physical Intelligence and FieldAI, both framed as building general-purpose model layers for robots rather than single-application software [The Robot Report, retrieved 2026]. That does not establish TAM, SAM, or SOM for nooma labs itself, and no named third-party market model in the supplied sources quantifies a market size for physics foundation action models. It does, however, support a narrower inference: investors increasingly view embodied AI as a distinct market category inside robotics software, with enough perceived upside to fund platform bets before broad commercialization is settled [The Robot Report, retrieved 2026].
The demand drivers surfaced in the cited research are practical rather than rhetorical. The Robot Report's coverage of Physical Intelligence and FieldAI points to a market need for models that can bridge perception, control, and adaptation in real-world robotic settings, especially where scripted systems break under variability [The Robot Report, retrieved 2026]. That maps cleanly onto nooma labs' own product claim around "physics foundation action models for robotics," which suggests an attempt to encode physical dynamics at the model level rather than treat control as a downstream engineering problem [nooma labs, retrieved 2024]. In market terms, the tailwind is not simply more robots, it is a growing willingness to spend on software abstractions that could reduce task-specific retraining and improve transfer across environments, if performance proves out [The Robot Report, retrieved 2026].
Adjacent markets matter because nooma labs appears to sit between academic research, model infrastructure, and applied robotics software. On one side is the broader robot foundation model stack, where The Robot Report highlights open-source and proprietary approaches, including Physical Intelligence's Pi0 release and a set of infrastructure platforms shaping physical AI in 2026 [The Robot Report, retrieved 2026]. On the other side are substitute approaches that rely on narrower control systems, classic robotics planning, or application-specific models, which remain the incumbent path in many deployments by implication, since generalized embodied models are still described as an emerging category rather than an established standard [The Robot Report, retrieved 2026]. For nooma labs, that means the relevant market is not only "robotics" in the abstract, but the contested layer between research-grade embodied intelligence and production-grade robotic autonomy.
Regulatory and macro forces are supportive in a broad sense but not yet company-defining on the public record. The supplied sources do not identify a nooma labs-specific regulatory posture, customer segment, or jurisdictional go-to-market motion. What they do show is that capital markets remain open for physical AI narratives, with large financings reported for category leaders, and that the field is still shaped by research institutions and independent labs as much as by standard enterprise software buyers [The Robot Report, retrieved 2026] [crosslabs.org, retrieved 2026]. For an early laboratory positioning itself in Kyoto and linking its work to foundational questions of agency, cognition, and adaptation, the near-term market question is less about regulation than about whether the category moves from research enthusiasm to repeatable commercial adoption in robotics programs with real budgets [NOOMA Labs, The Space Between Minds, retrieved 2026].
| Analogous market signal | Claim | Relevance to nooma labs |
|---|---|---|
| Physical Intelligence financing | Reported $400M raise for foundation models for robotics [The Robot Report, retrieved 2026] | Indicates investor appetite for general-purpose robotics model platforms |
| Physical Intelligence financing | Reported $600M raise to advance robot foundation models [The Robot Report, retrieved 2026] | Suggests continued category expansion rather than a one-off financing event |
| FieldAI financing | Reported $405M raise to scale "physics first" foundation models for robots [The Robot Report, retrieved 2026] | Most directly analogous public signal for nooma labs' physics-oriented positioning |
| Category formation | The Robot Report identified five physical AI infrastructure platforms shaping robotics in 2026 [The Robot Report, retrieved 2026] | Supports the view that infrastructure for embodied AI is emerging as a discrete market layer |
The table does not size the market, but it does show where outside capital and media attention are concentrating: model-centric infrastructure for robotics. For nooma labs, the investable question is whether that attention broadens into a durable software market, or remains concentrated among a small set of well-funded research-led platforms.
No independent source found -- This section relies on company websites and category-level trade reporting, with no independently corroborated market sizing for nooma labs or for physics foundation action models as a standalone market.
The Competitive Field
Competitive Map
MIXED The company appears to be positioning itself at the research-heavy end of robotics foundation models, closer to labs building core model and control abstractions than to application-layer robotics software vendors, but the public record around nooma labs itself is still thin [noomalabs.com] [crosslabs.org].
On the public evidence, the nearest competitive frame is not a list of directly confirmed rivals but a set of better-capitalized companies pursuing adjacent versions of robot foundation models and physical AI infrastructure. The Robot Report identifies Physical Intelligence as a company advancing foundation models for robotics and reports financings of $400 million and later $600 million, alongside product work such as the open-sourcing of its Pi0 robotics foundation model [The Robot Report]. The same publication reports that FieldAI raised $405 million to scale what it described as "physics first" foundation models for robots [The Robot Report]. That places nooma labs, whose site describes "physics foundation action models for robotics," in a field where a few named players already have substantial capital formation and clearer public visibility [noomalabs.com] [The Robot Report].
A second segment sits adjacent rather than directly head-on: infrastructure providers and research platforms that help train, evaluate, or deploy physical AI systems without necessarily owning the end model category. The Robot Report's overview of physical AI infrastructure platforms suggests an ecosystem in which model builders can be disaggregated from tooling, simulation, or deployment layers [The Robot Report]. If that architecture persists, nooma labs could compete for mindshare and technical credibility at the model layer while relying on outside infrastructure for scale, though that also leaves room for integrated competitors to compress margins and development cycles [The Robot Report] [noomalabs.com].
The third substitute set is conventional robotics stacks that do not rely on foundation-model-style approaches at all. The available sources do not name specific incumbents in motion planning, controls, or vertical robotics applications for direct comparison, so the cleaner inference is narrower: any customer or research partner evaluating nooma labs is also evaluating whether a foundation-action-model approach improves enough on task-specific robotics software to justify model risk, data collection cost, and longer validation cycles [noomalabs.com]. That is a meaningful competitive burden because the company's public materials establish ambition, but not yet deployments, benchmarks, or commercial proofs that would separate research novelty from production utility [noomalabs.com].
Edge and Exposure
MIXED The most plausible edge visible today is intellectual positioning rather than distribution. Public materials describe nooma labs as an independent research laboratory in Kyoto, and Cross Labs archive material says the research program developed and led by Olaf Witkowski continues at Nooma Labs, while also stating that the entity is independent and not involved with Cross Compass Ltd. [crosslabs.org] [noomalabs.org] [olafwitkowski.com]. If that continuity reflects a real transfer of scientific agenda and collaborators, the edge is a coherent research thesis that links agency, cognition, adaptation, learning, collective intelligence, and computation, which is broader than a narrow product claim and could attract technical talent or academic partnerships [noomalabs.org] [crosslabs.org].
That edge is also perishable. Research lineage can help with early recruiting and intellectual distinctiveness, but it does not by itself create a durable moat against competitors with deeper compute budgets, larger robotics data pipelines, or existing hardware relationships. Physical Intelligence and FieldAI already have named funding scale in the hundreds of millions according to The Robot Report, which usually translates into faster dataset acquisition, larger teams, and more room for iteration across embodiments [The Robot Report]. Until nooma labs shows public evidence of proprietary data access, hardware distribution, or benchmark leadership, its current edge reads as talent-and-thesis driven rather than structurally defended [noomalabs.com] [crosslabs.org].
The main exposure is straightforward: capitalized peers can define the category before nooma labs defines itself. Physical Intelligence has the clearest public advantage because it pairs large financings with visible model releases, giving it both narrative control and a route to ecosystem adoption [The Robot Report]. FieldAI's reported "physics first" framing also narrows the whitespace around nooma labs' own positioning, because it competes on similar conceptual terrain while entering the market with clearer funding validation [The Robot Report]. Nooma labs is also exposed to adjacent infrastructure companies if customers prefer to buy picks-and-shovels tooling around physical AI rather than commit to one early model provider [The Robot Report].
Likely 18-Month Scenario
MIXED The most plausible 18-month scenario is a sorting of the field into three buckets: a few heavily financed foundation-model leaders, a broader tooling layer, and a long tail of research labs that either prove a distinctive capability or remain primarily scientific organizations. Winner if X: Physical Intelligence, if open model distribution and large-scale financing continue to compound into developer adoption and hardware partnerships [The Robot Report]. Loser if Y: nooma labs, if it remains publicly legible as a research program without corresponding evidence of deployment, benchmark performance, or commercialization milestones that make the thesis investable beyond the lab context [noomalabs.com] [crosslabs.org].
There is still an upside case for nooma labs within that same period. If the company can translate its Kyoto-based research identity and Cross Labs continuity into a clearly differentiated action-model architecture, especially one that performs across multiple robotic settings rather than a single task family, it could occupy a more distinct niche than its current sparse public footprint suggests [noomalabs.org] [crosslabs.org] [noomalabs.com]. At this stage, though, the competitive landscape is defined less by direct feature comparison and more by asymmetry in proof: peers such as Physical Intelligence and FieldAI already have public financing and category coverage, while nooma labs has only a high-level product claim and research lineage on the record [The Robot Report] [noomalabs.com] [crosslabs.org].
Opportunity
PUBLIC
The prize here is unusually large because a company that helps robots act reliably in the physical world can sit in one of the hardest and most valuable control points in applied AI, assuming the underlying models prove transferable beyond a research setting [noomalabs.com] [The Robot Report].
The headline opportunity is to become a specialized model layer for robotics, not a general AI lab with diffuse ambitions. That framing follows from the only direct company statement currently visible, which says nooma labs is building "physics foundation action models for robotics" [noomalabs.com]. The broader category is no longer theoretical. The Robot Report has documented substantial capital formation around robotics foundation models, including Physical Intelligence's $400 million raise and later $600 million raise, as well as FieldAI's $405 million raise around a "physics first" approach [The Robot Report]. That does not validate nooma labs' execution, but it does establish that investors and strategic buyers now treat embodied AI as a serious infrastructure market rather than a lab curiosity [The Robot Report].
The upside paths are still narrow, but they are legible enough to outline.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Robotics foundation model supplier | nooma labs develops a model stack that other robotics companies license or build on for manipulation, planning, or control in physics-rich settings | A public research release or partner validation that shows the models generalize across tasks or hardware | The company's own positioning is explicitly around "physics foundation action models for robotics," and third-party coverage shows growing demand for foundation-model infrastructure in robotics [noomalabs.com] [The Robot Report] |
| Independent research lab to commercial bridge | nooma labs converts a research identity into a commercialization path, becoming the applied outlet for a longer-running scientific program around agency, cognition, and adaptation | Demonstrated translation of prior research into deployable robotics systems or tools | nooma labs is described by related archival sources as an independent Kyoto laboratory that continues the scientific direction of Cross Labs and the research program led by Olaf Witkowski [crosslabs.org, August 2026] [crosslabs.org] |
| Physical AI platform acquisition target | nooma labs becomes strategically valuable to a larger robotics, industrial, or AI platform seeking in-house embodied model capability | Category consolidation after more benchmark progress and capital concentration in embodied AI | Large financings for Physical Intelligence and FieldAI suggest strategic value is accruing to teams building the model layer for robots, which can pull smaller technically differentiated labs into acquisition discussions if they show credible results [The Robot Report] |
The table points to the same underlying issue: the company does not need to win all of robotics to matter. If it can own a difficult slice of robot action modeling, especially one that improves transfer across environments or hardware, the business could become upstream infrastructure for a much wider set of operators [noomalabs.com] [The Robot Report].
What compounding looks like in this category is a research-to-data-to-performance loop. Better action models attract more experiments, deployments, and partner interest. Those, in turn, can generate broader task data and failure-case insight, which should improve model performance if the training and evaluation stack is sound. Public evidence that this loop has already started at nooma labs is limited, but the company's narrow product statement and the archival continuity from Cross Labs suggest a concentrated research agenda rather than a broad product sprawl [noomalabs.com] [crosslabs.org, August 2026]. In embodied AI, focus itself can be an early advantage because the hard part is not feature breadth, it is repeated gains against real-world physical constraints [The Robot Report].
The size of the win is easiest to frame through category comparables rather than company fundamentals, because no revenue, customer, or funding data are publicly confirmed for nooma labs. Physical Intelligence has raised $400 million and later $600 million for robotics foundation model work, while FieldAI has raised $405 million around a related physics-first thesis [The Robot Report]. Those figures are financing amounts, not valuations, and they should not be mapped directly onto this company. Even so, they anchor the upper end of market appetite. If nooma labs were to establish credible technical leadership in a commercially relevant robotics niche, the outcome could plausibly be a strategic asset worth hundreds of millions of dollars (scenario, not a forecast), either as an independent model platform or as an acquisition target for a larger robotics or AI stack owner [The Robot Report].
One source, partially checked -- This section relies on one direct company source for positioning and one independent publisher for category validation, with archival corroboration from Cross Labs for research continuity.
Sources
Publicly reported
[noomalabs.com, 2024] nooma labs | https://www.noomalabs.com/
[The Robot Report, 2026] Physical Intelligence raises $600M to advance robot foundation models | https://www.therobotreport.com/physical-intelligence-raises-600m-advance-robot-foundation-models/
[NOOMA Labs, 2026] Nooma Labs | https://noomalabs.org/
[crosslabs.org, 2026] About Cross Labs | Historical Research Archive | https://www.crosslabs.org/team
[crosslabs.org, August 2026] A New Chapter: Nooma Labs | Cross Labs Archive | https://www.crosslabs.org/nooma-labs
[olafwitkowski.com, 2026] Dr. Olaf Witkowski | Unlocking Artificial Collective Intelligence | https://olafwitkowski.com/
[The Robot Report, 2026] Physical Intelligence raises $400M for foundation models for robotics | https://www.therobotreport.com/physical-intelligence-raises-400m-for-foundation-models-for-robotics/
[The Robot Report, 2026] Physical Intelligence open-sources Pi0 robotics foundation model | https://www.therobotreport.com/physical-intelligence-open-sources-pi0-robotics-foundation-model/
[The Robot Report, 2026] FieldAI raises $405M to scale 'physics first' foundation models for robots | https://www.therobotreport.com/fieldai-raises-405m-scales-physics-first-foundation-models-robots/
[The Robot Report, 2026] 5 Physical AI infrastructure platforms shaping robotics in 2026 | https://www.therobotreport.com/5-physical-ai-infrastructure-platforms-shaping-robotics-in-2026/
Articles about nooma labs
- Kyoto's Nooma Labs Is Betting on Physics for Robot Intelligence — The independent Kyoto research laboratory continues a scientific program from Cross Labs, aiming to connect foundational questions about cognition with new approaches to learning.