RoboScience

Develops general-purpose embodied AI, including robot-intelligence modules and full robot systems like REX G1.

Cover Block

Publicly reported

Field	Detail
Company	RoboScience
Tagline	Develops general-purpose embodied AI, including robot-intelligence modules and full robot systems like REX G1 [RoboScience]
Headquarters	Beijing, China [Crunchbase]
Founded	2024 [Crunchbase]
Stage	Series A [Preqin, May 2026]
Business model	Hardware + Software
Industry	Deeptech
Technology	AI / Machine Learning
Geography	East Asia
Growth profile	Venture Scale
Founding team	Co-Founders (3+) [RoboScience] [LinkedIn / Jingbo Liu] [LinkedIn / Xiaojin Shi]
Funding label	$100M+
Total disclosed	CNY 1 billion Series A reported by Preqin, with earlier and later round amounts not consistently disclosed across public sources [Preqin, May 2026]

Summary and Signal

PUBLIC RoboScience is a Beijing-based embodied AI startup building both robot intelligence modules and full robot systems, and it merits attention because it has moved from founding in late 2024 to a reported Series A financing path by 2026 in one of the more capital-intensive corners of applied AI [RoboScience] [Preqin, May 2026] [PANews, November 2026]. The public record is still thin in places, but the pace of financing and the company’s attempt to pair model development with deployable hardware make it more than a pure research story [PANews, November 2026] [Dealroom].

The founding story, as publicly described, centers on Tian Ye and a broader founding group that includes Shao Lin, Liu Penghai, Wang Tao, Jingbo Liu, and Xiaojin Shi, with the company stating it was founded in December 2024 [RoboScience] [LinkedIn / Jingbo Liu] [LinkedIn / Xiaojin Shi]. RoboScience presents itself as pursuing cross-embodiment general-purpose embodied intelligence, meaning software intended to transfer manipulation capability across different robot forms, objects, and task settings rather than solving for a single machine or workflow [Dealroom] [RoboScience].

Its product stack is described in layered terms: the Visics embodied AI model, the RoboMirage simulation engine, cloud-based EaaS, and the REX series of wheeled humanoid robots, including REX G1 [RoboScience] [Pandaily, June 2026]. The differentiator, if the company can validate it in the field, is not simply that it builds robots, but that it claims a generalizable control and world-model layer that can operate across hardware embodiments, a harder and more valuable proposition than a single-purpose automation product [Dealroom] [Pandaily, June 2026].

The team profile is one reason investors have engaged early. Public materials describe Tian Ye as having Stanford AI training and prior large-scale AI infrastructure experience, while Shao Lin is listed as an assistant professor at the National University of Singapore, and Liu Penghai and Wang Tao are positioned around hardware operations and capital formation respectively [National University of Singapore] [LinkedIn / 01VC, August 2025] [RoboScience]. Some biography details rely on company or investor-linked materials rather than broad independent reporting, so the backgrounds are directionally relevant but not fully triangulated [RoboScience] [LinkedIn / 01VC, August 2025].

Funding momentum has been substantial on paper. PANews reports seed, angel, Pre-A, and Series A2 financings across 2025 and 2026, while Preqin lists a Series A round in May 2026; the company is associated with investors including 01VC, JD.com, China Merchants Venture, SenseTime Guoxiang Capital, Puhua Capital, Fortune Capital, and Amber Capital [PANews, November 2026] [Preqin, May 2026]. The business model appears to combine hardware sales with software and embodied-intelligence licensing, but public sources do not yet establish named customers or verified deployment scale, which is the main gap to watch [Dealroom] [RoboScience] [PANews, November 2026].

Over the next 12 to 18 months, the key question is whether RoboScience can convert technical ambition and financing breadth into repeatable commercial evidence. Investors should watch for independently confirmed deployments, clearer proof that the model layer works across multiple robot embodiments, and signs that economics can support adoption in industrial or logistics settings where embodied AI still faces real engineering, safety, and maintenance constraints [iTechGuides, 2026] [Healthcare, March 2026] [SAE World Congress 2026].

No independent source found -- This section relies materially on company descriptions and investor-linked or single-source reporting for product scope, founder biographies, and financing chronology, with partial corroboration from Dealroom, Preqin, and NUS.

Taxonomy Snapshot

Axis	Value
Stage	Series A
Business Model	Hardware + Software
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	East Asia
Growth Profile	Venture Scale
Founding Team	Co-Founders (3+)
Funding	$100M+

Company Overview

PUBLIC

RoboScience presents as a very young Beijing embodied-AI company, with the public record anchored more by company profile data than by formal corporate filings in the supplied materials. Crunchbase lists the company as founded in 2024 and headquartered in Beijing, China, while the company website states a more specific founding date of December 2024 [Crunchbase] [RoboScience]. That dating matters because it compresses the operating history to less than two years at the time of the latest cited financing, which makes chronology more informative than scale claims in assessing where the business stands [Crunchbase] [RoboScience].

The company describes its core work as research and development in cross-embodiment general-purpose embodied intelligence, with products spanning robot-intelligence modules, full robot systems, and the REX line [RoboScience]. In practice, the public description suggests a dual ambition: building model infrastructure that can generalize across hardware formats, and commercializing complete robotic systems rather than software alone [RoboScience].

The early milestones visible from public company-linked sources are straightforward. The business was founded in December 2024 according to the company website, and by the time of its current public materials it was presenting a named stack that includes the Visics model, the RoboMirage simulation engine, and the REX series, alongside what it calls an Embodied-AI-as-a-Service offering [RoboScience]. Those are product-positioning milestones rather than independently verified commercial milestones, but they do show how quickly the company moved from formation to a full platform narrative [RoboScience].

One source, partially checked -- Confirmed by Crunchbase and the company website; legal entity details and operating milestones remain primarily company-sourced.

The Product and the Stack

MIXED RoboScience is presenting itself as a full-stack embodied AI company rather than a point solution. Public company materials say it develops "general-purpose embodied AI," spanning robot-intelligence modules, full robot systems, the REX series of wheeled humanoid robots, the Visics model, and a simulation engine called RoboMirage [RoboScience]. Dealroom describes the underlying technical aim more plainly: cross-embodiment embodied intelligence that can carry manipulation skills across different hardware, objects, and operating scenarios [Dealroom].

The most specific external description of the model layer comes from Pandaily, which reports that Visics uses a dual-engine architecture: an Embodied World Model trained on internet video data to learn object physics, motion patterns, and force dynamics, and a General Operation Model that converts those understandings into executable object trajectories [Pandaily, June 2026]. RoboScience also says it has launched a cloud-based Embodied-AI-as-a-Service offering, or EaaS, alongside the REX product line, but the public record provided here does not independently verify usage, customer adoption, or benchmark performance against other embodied AI stacks [RoboScience].

The engineering footprint implied by hiring is directionally consistent with that positioning, but it should still be treated as inference rather than proof of shipped capability. Publicly surfaced roles include reinforcement learning, algorithm design and development, and "Generative AI Models and World Models for Robot Learning," which suggests active investment in model training, simulation, and control systems (inferred from job postings) [RoboScience] [freehire]. What is still missing in public is the operating evidence that matters most for an embodied AI claim: verified demos under production conditions, named customer deployments, or repeatable task-performance data across multiple robot embodiments [PANews, November 2026].

No independent source found -- Material product claims rely primarily on RoboScience's own website, with limited third-party technical detail from Dealroom and Pandaily.

The Market They Are Entering

PUBLIC The market matters now because embodied AI has moved from laboratory promise to early field deployment, but the evidence still points to a constrained adoption curve shaped as much by engineering economics and safety discipline as by model progress [iTechGuides, 2026] [SAE World Congress 2026].

RoboScience is positioned inside the embodied AI and intelligent robotics stack, yet the available public record does not provide a named third-party TAM, SAM, or SOM specific to the company. In that absence, the cleaner approach is to frame the addressable market through adjacent public categories rather than force unsupported sizing. Public descriptions place the company across industrial, logistics, retail, and household use cases, with a model that combines robot hardware, AI modules, and cloud software, which suggests exposure to both the robotics automation budget and the emerging spend bucket for embodied AI tooling rather than a single narrow end market [RoboScience] [Dealroom].

The demand signal in current research is real, but narrower than some category rhetoric implies. An industry review cited in the source set argues that the strongest 2026 deployments show embodied AI can add a flexible tool to selected workflows, not that a general-purpose robot workforce has arrived [iTechGuides, 2026]. That matters for market formation: buyers appear more likely to start with bounded, repetitive environments where manipulation and mobility can be measured against labor availability, uptime, and safety constraints, especially in logistics and industrial settings that already purchase automation.

The main tailwind is technical convergence. RoboScience's own positioning rests on cross-embodiment operation, meaning the same intelligence layer can in principle transfer across different robot hardware and task settings, and outside commentary in the source set describes the category's technical wedge in similar terms [Dealroom] [LinkedIn, March 2025]. If that transferability holds in practice, the commercial implication is a broader software surface per deployed robot platform, though the public evidence here does not yet establish customer conversion or repeat deployments.

The constraints are as important as the tailwinds. A 2026 SAE World Congress assessment argues that successful embodied AI adoption will depend on pairing capability gains with engineering discipline, safety thinking, and governance, which is a reminder that market expansion is not just a model-quality story [SAE World Congress 2026]. Separate healthcare-focused research in the source set reaches a similar conclusion from a different angle, noting that high acquisition and maintenance costs and limited economic evidence remain material barriers to adoption decisions, especially in sectors with tighter procurement and compliance requirements [Healthcare, March 2026].

Adjacent markets also shape the opportunity set. If fully generalist embodied systems remain expensive or operationally immature, buyers can continue to spend on substitutes such as fixed-function industrial automation, warehouse robotics, software-only computer vision, and conventional labor outsourcing. That does not cap the category, but it does raise the proof threshold for startups selling general-purpose systems: the product must outperform established automation options on task breadth, integration effort, or economics, and ideally on all three over time [iTechGuides, 2026].

The regulatory backdrop is still forming, but the broad direction is visible. Research cited here emphasizes safety and governance as prerequisites for scaled adoption, which likely means longer enterprise validation cycles and tighter scrutiny in environments where robots operate near workers or consumers [SAE World Congress 2026]. For a company with both hardware and AI layers, that can slow rollout velocity while also favoring teams that can show disciplined system engineering rather than model performance in isolation.

Market lens	Public claim	Relevance to RoboScience
Embodied AI deployment maturity	The strongest 2026 deployments support selected workflows rather than a broad general-purpose robot workforce [iTechGuides, 2026]	Suggests near-term demand is likely task-specific before platform-wide expansion
Safety and governance	Adoption depends on AI capability plus engineering discipline, safety thinking, and governance [SAE World Congress 2026]	Points to longer validation cycles and higher system-integration burden
Healthcare as a proxy for regulated adoption	High acquisition and maintenance costs, plus limited economic evidence, constrain adoption [Healthcare, March 2026]	Useful analog for any vertical where ROI and compliance scrutiny are high

The table points to a market that is opening, but unevenly. The practical read is that RoboScience is pursuing a large category in concept, while the monetizable market in the next few years is more likely to be built one constrained workflow and one validated vertical at a time.

One source, partially checked -- Market context in this section relies primarily on three named public sources, with no company-specific third-party TAM dataset in the provided materials.

The Competitive Field

MIXED RoboScience is positioning itself less as a single robot vendor than as a cross-embodiment intelligence stack, which places it in competition with integrated robotics companies on one side and model-layer or simulation-layer alternatives on the other [Dealroom] [RoboScience].

The public record is thin on named direct rivals, and that matters for how this section should be read. Available sources consistently describe RoboScience as building general-purpose embodied AI, proprietary model infrastructure, and full robot systems such as the REX line, while also framing a business model that includes licensing AI modules to robotics manufacturers and automation enterprises [Dealroom] [RoboScience] [RobotHub]. That suggests three practical competitive sets: integrated robot makers that sell complete systems into real-world workflows, AI-first embodied intelligence companies that aim to generalize across hardware, and adjacent substitutes such as traditional industrial automation or task-specific service robots that may solve narrower use cases with better operating economics [Dealroom] [Healthcare, March 2026] [iTechGuides, 2026].

Within those sets, the nearest incumbent substitute is not necessarily another general-purpose humanoid program. In sectors such as logistics, retail, and industrial operations, buyers can still choose fixed automation, warehouse software, or purpose-built robots over a broader embodied AI platform if reliability and payback are clearer [Healthcare, March 2026] [iTechGuides, 2026]. That creates a familiar commercialization problem for early embodied AI companies: the technology story can be broader than the procurement story. RoboScience's own positioning toward industrial, logistics, retail, and household scenarios gives it a large addressable field, but it also means it is confronting established buyer behavior in each segment rather than a single clean category [RoboScience] [Dealroom].

RoboScience's clearest edge today appears to be team composition and capital access, though both are better evidenced than commercial traction. Public materials describe a founding group spanning AI research, hardware and supply chain experience, and fundraising networks, including Tian Ye, Shao Lin, Liu Penghai, and Wang Tao [National University of Singapore] [RoboScience] [LinkedIn / 01VC, August 2025]. The financing cadence is also unusually fast for a company founded in late 2024, with seed, angel, Pre-A, Series A, and Series A2 rounds all reported within roughly two years, and investors that include JD.com, 01VC, Puhua Capital, Amber Capital, China Merchants Venture, SenseTime Guoxiang Capital, and others [PANews, November 2026] [Preqin, May 2026] [LinkedIn / 01VC, August 2025]. That can be durable if it translates into training infrastructure, hiring, and field deployments before peers do. It is perishable if the company remains long on technical ambition and short on named customers, because capital is an input, not a moat.

The main exposure is that several claimed differentiators are still company-described rather than independently validated. RoboScience says it has built the Visics embodied AI model, the RoboMirage simulation engine, and an "Embodied-AI-as-a-Service" offering, and PANews also reported technical details of a dual-engine architecture for Visics [RoboScience] [Pandaily, June 2026]. But the same public record does not identify named paying customers, deployment counts, or procurement wins that would show these components outperform adjacent options in live environments [PANews, November 2026]. In practical terms, that leaves the company vulnerable to any better-capitalized robotics platform or established automation vendor that can show lower integration risk, stronger service coverage, or clearer economics in a narrow vertical. It also leaves RoboScience exposed to a category boundary of its own making: if customers buy outcomes rather than generality, broad cross-embodiment capability may not be the feature that closes the deal [iTechGuides, 2026] [Healthcare, March 2026].

The most plausible 18-month scenario is not a single winner-take-all outcome but a sorting process between platforms that can prove repeatable deployments and those that remain primarily technical narratives. If embodied AI adoption continues to progress as a workflow tool rather than a general labor replacement, JD.com looks like a potential winner if it converts from investor to ecosystem anchor, because logistics and retail environments offer concrete testing grounds for robotics adoption, even though no public source here confirms a commercial partnership with RoboScience [PANews, November 2026]. The more exposed party, if the category slows further, is any early embodied AI vendor whose public momentum is driven mainly by fundraising headlines rather than operating evidence. On the available record, RoboScience has enough capital and senior talent to stay in the front pack, but the next competitive proof point needs to be customer evidence, not another architecture claim [PANews, November 2026] [RoboScience] [iTechGuides, 2026].

One source, partially checked -- Based mainly on company materials, Dealroom, PANews, Preqin, and sector commentary; no named competitors or independently verified customer benchmarks were available in the provided public sources.

Opportunity

Upside case

PUBLIC The size of the prize here is unusually large because the company is not trying to sell a single robot for a single workflow, it is trying to supply the model, simulation, and hardware stack that could sit underneath many robot form factors if embodied AI matures into a real systems market rather than a narrow product niche [Dealroom] [RoboScience] [Pandaily, June 2026].

The cleanest upside case is that RoboScience becomes a category platform for cross-embodiment manipulation, meaning its software and training stack are used across multiple robot designs while its own REX systems serve as reference hardware and early commercialization vehicles [Dealroom] [RoboScience] [Pandaily, June 2026]. That outcome is still speculative, but it is at least tethered to public evidence. The company describes itself not as a single-task robotics vendor but as a builder of "general-purpose embodied AI," with a proprietary Visics model, a simulation engine called RoboMirage, and cloud-based EaaS alongside full robot systems [RoboScience] [Pandaily, June 2026]. The financing record also matters here. Across seed, angel, Pre-A, Series A, and Series A2, public reports show repeat backing from venture and strategic investors including 01VC, JD.com, China Merchants Venture, SenseTime Guoxiang Capital, Puhua Capital, Fortune Capital, Amber Capital, and Shanghai Semiconductor Industry Investment, which suggests the market is funding the platform ambition rather than a one-off prototype effort [PANews, November 2026] [Preqin, May 2026] [LinkedIn / 01VC, August 2025].

The near-term reason this is reachable rather than purely aspirational is team composition. Public materials point to a mix of AI research, hardware operations, and capital-markets experience: Tian Ye is described in investor and company-linked materials as having Stanford AI training and prior AI-platform experience, Shao Lin is listed by NUS as an assistant professor, and Liu Penghai is presented by the company as a former Ecovacs executive with hardware and supply-chain experience [National University of Singapore] [LinkedIn / 01VC, August 2025] [RoboScience]. That does not prove execution, but it does fit the technical and operational demands of embodied AI better than a software-only founding bench would.

Scenario	What happens	Catalyst	Why it's plausible
Reference-stack winner	RoboScience's Visics model, RoboMirage simulator, and EaaS layer become the preferred development stack for robotics manufacturers that need manipulation intelligence faster than they can build it internally	A strategic investor or manufacturing partner converts from financier to integration channel, turning the software stack into embedded infrastructure [PANews, November 2026] [LinkedIn / 01VC, August 2025]	The company already frames its business around licensing AI models and selling embodied-intelligence solutions across hardware platforms rather than only shipping one robot [Dealroom] [RoboScience]
Vertical beachhead to platform	The REX line gains traction in a narrow set of commercial workflows, then the data and deployment learnings feed back into broader model performance and more use cases	A first repeatable deployment category in logistics, retail, or industrial environments where manipulation quality matters more than humanoid form factor [RoboScience] [iTechGuides, 2026]	Public descriptions identify those sectors as target markets, and external commentary on embodied AI in 2026 suggests selected workflow adoption is the credible entry point, not universal labor replacement [RoboScience] [iTechGuides, 2026]
China-origin embodied AI champion	RoboScience becomes one of the better-capitalized independent embodied AI platforms in China, using funding depth and domestic investor networks to assemble a faster commercialization loop than smaller peers	Continued syndication from strategic and financial investors after the November 2026 Series A2, especially if semiconductor and industrial backers deepen involvement [PANews, November 2026] [TechStartups, November 2026]	The cap table already includes a mix of platform, industrial, and financial investors, and PANews reports a multi-round financing history in less than two years [PANews, November 2026] [Preqin, May 2026]

The compounding logic is straightforward, even if the proof is still early. If RoboScience can put one intelligence layer across multiple embodiments, each deployment can generate additional task data, simulation feedback, and failure cases that improve the general model and make the next customer cheaper to serve and faster to onboard [Dealroom] [RoboScience] [Pandaily, June 2026]. The simulator matters in that loop because a differentiable multimodal physical simulation engine, if it works as described, could let the company train and validate policies before costly real-world deployment, tightening iteration speed across robot types [RoboScience]. Public signals that this flywheel is beginning are modest rather than decisive: the company is hiring for reinforcement learning and world-model-related roles, and its public materials consistently present software, simulation, and hardware as one stack rather than disconnected products [RoboScience] [freehire] [RoboScience, retrieved 2026].

The size of the win is best framed as a scenario, not a forecast. The public sources provided here do not include a clean third-party TAM estimate specific to cross-embodiment embodied AI, and the external research is more cautionary than promotional, stressing that adoption still depends on engineering discipline, safety, governance, and clearer economic evidence in many sectors [SAE World Congress 2026] [Healthcare, March 2026]. Even so, if the "reference-stack winner" scenario plays out, the value could resemble a full-stack robotics platform rather than a conventional component supplier. Preqin records a Series A round of CNY 1 billion in May 2026, and PANews describes a further Series A2 in November 2026 worth "hundreds of millions of yuan," which places RoboScience in the financing range of companies being capitalized for category-scale outcomes rather than niche automation projects [Preqin, May 2026] [PANews, November 2026]. A reasonable interpretation is that, under a successful platform scenario, this could become a multibillion-dollar company (scenario, not a forecast), provided it proves repeatable deployments and that its model layer travels across hardware and tasks. The public evidence supports the existence of that ambition and the resourcing behind it. It does not yet prove the market has validated it.

One source, partially checked -- This section relies on one independent financing database, one named media report, and company materials for product architecture and commercialization claims [Preqin, May 2026] [PANews, November 2026] [RoboScience].

Sources