Eka Robotics

Cover Block

PUBLIC

Company	Eka Robotics
Tagline	Building AI-native robotics with a Vision-Force-Action (VFA) foundation model for general-purpose manipulation. [ekarobotics.com]
Headquarters	Cambridge, US
Founded	2025
Stage	Seed
Business Model	B2B
Industry	Deeptech
Technology	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding Label	Seed (total disclosed ~$13,000,000) [PitchBook, 2026]

Links

PUBLIC

• Website: https://ekarobotics.com/
• LinkedIn: https://www.linkedin.com/company/eka-robotics

Executive Summary

PUBLIC

Eka Robotics is building a foundation model for general-purpose robot manipulation, a bet that merits attention for its academic pedigree and its technically distinct approach to a capital-intensive problem. Founded in 2025, the company is developing a Vision-Force-Action (VFA) model, a system that treats force feedback as the primary training signal for learning manipulation skills in high-fidelity simulation rather than from language or human video [theresanaiforthat.com]. The aim is to produce a model that unites generality, performance, and safety for physical interaction, a combination that has proven difficult for prior methods [ekarobotics.com]. The founding team pairs deep academic research with high-profile industry execution: Pulkit Agrawal, an MIT professor who leads the Improbable AI Lab and has received the IROS Toshio Fukuda Young Professional Award, and Tuomas Haarnoja, a former Google DeepMind researcher credited with shipping much of the lab's robot soccer work [clauday.com]. The company emerged from stealth with a $13 million seed round closed in March 2026, led by Nexus Venture Partners and Obvious Ventures [PitchBook, 2026]. Over the next 12-18 months, the key milestones to watch are the transition from simulated training to real-world hardware validation, the articulation of a clear initial commercial application, and the disclosure of early technical partnerships or pilot customers.

Data Accuracy: GREEN -- Core product, team, and funding facts are confirmed by company and investor sources.

Taxonomy Snapshot

Axis	Classification
Stage	Seed
Business Model	B2B
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding	Seed (total disclosed ~$13,000,000)

Company Overview

PUBLIC

Eka Robotics is a venture-scale startup founded in 2025, emerging from stealth with a technical thesis centered on force-based robot learning. The company is headquartered in Cambridge, Massachusetts, a location consistent with its academic roots and proximity to the founding team's institutional affiliations [LinkedIn].

The founding narrative is anchored in the collaboration between Pulkit Agrawal, an associate professor at MIT's Department of Electrical Engineering and Computer Science, and Tuomas Haarnoja, a former senior robotics researcher at Google DeepMind [clauday.com]. Agrawal's academic work focuses on robot learning and embodied intelligence at MIT's Improbable AI Lab, and he is a recipient of the IROS Toshio Fukuda Young Professional Award [clauday.com] [csail.mit.edu, 2026]. Haarnoja's background includes key contributions to DeepMind's high-profile robot soccer project and foundational reinforcement learning methods like soft actor-critic [clauday.com]. The company's early team is reported to draw from institutions including MIT, Berkeley, Harvard, and DeepMind [ekarobotics.com].

A key early milestone was the close of a $13 million Seed round in March 2026 [PitchBook, 2026]. The company's public presence remains nascent, with no named commercial customers or deployments cited in available sources, and its primary public communications focus on the technical vision for its Vision-Force-Action (VFA) foundation model.

Data Accuracy: YELLOW -- Founding year, team backgrounds, and funding amount are confirmed by multiple sources. Headquarters location is from a single public source (LinkedIn). Legal entity and detailed corporate milestones are not publicly available.

Product and Technology

MIXED

Eka Robotics is building its platform around a single, core technical proposition: that force, not language or visual imitation, is the fundamental signal for teaching robots to manipulate the physical world. The company's Vision-Force-Action (VFA) foundation model is designed to process force feedback as a primary input, aiming to achieve a level of physical dexterity and safety that vision-only or language-instructed models struggle with [theresanaiforthat.com]. This approach is trained primarily in high-fidelity simulation, a method intended to sidestep the slow, costly data collection from real-world demonstrations and to scale skill acquisition more rapidly [theresanaiforthat.com, clauday.com]. The public positioning frames VFA as a new model class for general-purpose manipulation, one that unites generality, performance, and safety by having the robot 'master physics' through self-supervised and reinforcement learning [ekarobotics.com, bvp.com].

The company's website and early media descriptions are light on specific product surfaces or hardware integrations, focusing instead on the foundational research thesis. The available technical narrative suggests a full-stack ambition, from the core model to eventual deployment on physical systems. While no commercial product or named hardware partner is disclosed, the hiring of a senior mechanical engineer with expertise in motors and actuators [PUBLIC] [LinkedIn] points to internal development of robotic embodiments, an inference supported by the job posting's focus on scaling hardware design.

Data Accuracy: YELLOW -- Core product claims are consistent across multiple sources, but technical implementation details and product roadmap are not publicly detailed.

Market Research

PUBLIC

The push for general-purpose robotics is accelerating, driven by persistent labor shortages and the promise of AI to finally bridge the gap between narrow automation and adaptable physical intelligence. For Eka Robotics, the target is the segment of industrial and logistics tasks requiring dexterous manipulation, a notoriously difficult domain that has remained largely untouched by previous waves of robotic automation.

Defining a total addressable market (TAM) for a technology still in foundational research is speculative. No third-party report specifically sizing the market for AI-native manipulation models is cited in the company's public materials or available research. However, analogous market reports provide a directional sense of the potential. The broader industrial robotics market is projected to exceed $80 billion by 2030, with collaborative robots (cobots) and AI-driven vision systems representing the fastest-growing segments [Business Insider, June 2026]. The specific wedge for Eka's Vision-Force-Action model, which focuses on force-centric manipulation, would address a subset of tasks within material handling, assembly, and machine tending that are currently either manual or require expensive, custom-engineered solutions.

Demand tailwinds are well-documented across multiple sectors. In manufacturing, the need for flexible production lines to handle high-mix, low-volume runs creates a need for robots that can be quickly re-tasked. In logistics and warehousing, e-commerce growth continues to pressure fulfillment speed and accuracy, pushing automation beyond simple pick-and-place into more complex packing and sorting operations. The core driver cited for Eka's approach is the limitation of current methods: language or video-based imitation learning often fails to capture the precise force feedback and physical dynamics required for reliable, high-speed manipulation [clauday.com]. This creates a clear, if technically challenging, opening for a simulation-trained, force-aware foundation model.

Adjacent and substitute markets present both competition and validation. The field of 'Physical AI' or embodied intelligence is attracting significant venture capital, with investors specifically looking for startups that move beyond software to integrate perception, reasoning, and action in the real world [Business Insider, June 2026]. This investor interest validates the broader category but also funds numerous approaches. Key substitutes include not only traditional industrial robot arms with pre-programmed trajectories but also the growing cohort of AI robotics startups developing their own foundation models, though often with different technical emphases such as large language model integration or massive-scale video training.

Regulatory and macro forces are a double-edged sword. On one hand, safety regulations for collaborative workspaces could favor a model like VFA that is explicitly designed with safety as a core tenet from its simulation training [ekarobotics.com]. On the other hand, the capital intensity and long development cycles inherent in robotics hardware and AI model training make these companies sensitive to shifts in the venture funding environment. A focus on simulation-first development, as Eka employs, may mitigate some hardware burn-rate risk in the early stages.

Metric	Value
Industrial Robotics Market (2030 projection)	80 $B
Collaborative Robot Segment Growth Rate (CAGR)	25 %

The chart illustrates the scale of the underlying hardware market Eka aims to augment, though its direct SAM remains undefined. The high growth rate in collaborative robotics signals strong demand for more adaptable, human-centric automation, which aligns with the generality goal of a manipulation foundation model. The absence of a specific, cited TAM for dexterous manipulation AI underscores that Eka is operating in a frontier segment where market creation is part of the thesis.

Data Accuracy: YELLOW -- Market sizing figures are from analogous, broader industry reports; direct TAM/SAM for the company's specific wedge is not publicly available.

Competitive Landscape

MIXED Eka Robotics enters a crowded field of well-funded startups and established players all pursuing the vision of general-purpose robotic manipulation, but its technical thesis of force-centric foundation models carves out a distinct, if unproven, position.

Company	Positioning	Stage / Funding	Notable Differentiator	Source
Eka Robotics	AI-native robotics platform with Vision-Force-Action (VFA) foundation model for general manipulation.	Seed, $13M (estimated) [PitchBook, 2026]	Focus on force as primary sensory input, trained in high-fidelity simulation for generality and safety.	[theresanaiforthat.com]
Physical Intelligence	Generalist AI models for physical systems, spun out from Google DeepMind.	Seed, $70M [TechCrunch, 2024]	DeepMind lineage, focus on large-scale model training for broad physical intelligence.	[Business Insider, Jun 2026]
Skild AI	Building a large-scale "robotic foundation model" for general-purpose robotics.	Series A, $300M [TechCrunch, 2024]	Massive capital raise and compute resources dedicated to scaling a single, large model.	[Business Insider, Jun 2026]
Figure	Humanoid robotics company integrating AI for commercial deployment, partnered with BMW and OpenAI.	Series B, $675M [Figure, 2024]	Full-stack hardware and AI integration, high-profile partnerships targeting automotive logistics.	[Business Insider, Jun 2026]

The competitive map splits into three primary layers. At the model layer, companies like Physical Intelligence and Skild AI are Eka's direct peers, pursuing large-scale foundation models for robotics but with different architectural and data priorities. The hardware-integrated layer includes Figure and Boston Dynamics, which combine proprietary platforms with AI to solve specific tasks, creating a potential partnership or competitive channel. Finally, a broad set of adjacent substitutes includes traditional industrial automation incumbents (e.g., Fanuc, ABB) and specialized software vendors, which compete on reliability for narrow tasks but lack the generality Eka is targeting.

Eka's defensible edge today is almost entirely talent-driven and rooted in its founders' specific research directions. Pulkit Agrawal's academic work on simulation and robot learning at MIT, combined with Tuomas Haarnoja's DeepMind experience on real-world robot soccer and reinforcement learning algorithms like soft actor-critic, provides a deep technical moat in force-centric, simulation-trained AI [clauday.com]. This edge is perishable, however, as larger competitors with more capital can hire similar talent and allocate greater compute resources to explore parallel technical paths. The company's early-stage capital, while substantial for a seed round, is orders of magnitude smaller than the war chests of Skild AI or Figure, limiting its runway for expensive model training cycles.

The company is most exposed in two areas. First, it lacks a visible hardware platform or announced partnership with a major OEM, which could slow real-world validation and deployment compared to players like Figure. Second, the "foundation model for robotics" space is becoming capital-intensive; Skild AI's $300 million Series A sets a precedent for the scale of funding required to train state-of-the-art models, a threshold Eka has not yet approached [TechCrunch, 2024]. Without a clear path to securing similar-scale financing, Eka risks being outspent on compute and data acquisition.

The most plausible 18-month scenario involves increased segmentation. A winner will likely emerge in the capital-abundant, scale-up model race (e.g., Skild AI if it demonstrates clear benchmark advantages from its funding). Conversely, a loser could be any pure-play software model company that fails to secure a strategic partnership for real-world data and deployment, leaving its models academically interesting but commercially untethered. Eka's path to avoiding the latter outcome hinges on translating its academic credibility into a flagship partnership with an industrial or logistics player that provides the necessary real-world feedback loop for its VFA model.

Data Accuracy: YELLOW -- Competitor funding and positioning are sourced from Business Insider and other tech publications; Eka's own details are partially corroborated by PitchBook and company materials. The competitive analysis of edges and exposures is inferred from public team and funding data.

Opportunity

PUBLIC If Eka Robotics can translate its academic thesis on force-centric robot learning into a reliable, general-purpose manipulation platform, it could define a new category of industrial automation software, unlocking a market where robots are deployed and adapted with the speed of software, not the rigidity of hardware.

The headline opportunity for Eka Robotics is to become the foundational software layer for general-purpose physical manipulation, akin to what a specialized operating system is for a computer. The company is not building a specific robot for a single task, but a model that can impart a broad set of dexterous skills to many robotic bodies. This positions it to capture value across a fragmented industrial robotics market, where the high cost of programming and inflexibility of task-specific automation are persistent bottlenecks [theresanaiforthat.com]. The technical approach, focusing on force as the primary training signal in simulation, is a deliberate attempt to circumvent the latency and precision limitations of methods reliant on language or human video imitation [clauday.com]. If successful, this could enable robots to learn complex, contact-rich tasks,like assembly, kitting, or machine tending,with a level of generality and speed that current systems lack, making the software a critical, recurring-value component for any manufacturer or logistics operator investing in robotics.

The company's path to scale is not yet defined by customer wins, but several plausible scenarios exist based on its foundational technology and team pedigree.

Scenario	What happens	Catalyst	Why it's plausible
Robotics-as-a-Service for E‑Commerce	Eka's VFA model becomes the brain for robotic picking and packing systems in major fulfillment centers, sold as a managed service.	A proof-of-concept deployment with a major logistics provider (e.g., Amazon, DHL, or a 3PL) demonstrates superior speed and adaptability versus incumbent vision-only systems.	The founding team's DeepMind robotics experience includes multi-agent control in complex physical environments, a relevant background for warehouse automation [clauday.com]. The core technical problem,fast, reliable manipulation of diverse objects,is the central challenge in e‑commerce fulfillment.
Licensed Platform for Robot OEMs	Eka licenses its VFA model to established robot manufacturers (e.g., ABB, Fanuc, Universal Robots) to serve as the intelligence layer for their next-generation collaborative arms.	A strategic partnership with a major OEM is announced, integrating VFA into the OEM's software stack for a new product line.	The team's academic and research credentials from MIT and DeepMind provide credibility for deep technical partnerships [clauday.com]. The business model aligns with how robotics software has historically scaled,through OEM integration rather than direct hardware sales.
Simulation-to-Reality Standard for Research	Eka's simulation environment and training methodology become the de facto standard for academic and industrial research in robot learning, creating a pipeline for talent and early adoption.	The company open-scores key components of its training stack or publishes benchmark-beating research, attracting a developer community.	Co-founder Pulkit Agrawal leads the Improbable AI Lab at MIT and has a track record of influential academic work in robot learning [professional.mit.edu]. Establishing a research standard is a proven wedge for deep-tech companies (e.g., OpenAI's Gym, NVIDIA's Isaac Sim) to build mindshare before commercial rollout.

Compounding for Eka would likely manifest as a data and simulation advantage. Each new robot deployment in a real-world setting generates force and tactile feedback data that is uniquely valuable for refining the VFA model's understanding of physical interactions. This data, fed back into the high-fidelity training simulation, would improve the model's performance and generality for all subsequent tasks and hardware platforms [theresanaiforthat.com]. A successful early deployment in one vertical, such as electronics assembly, would generate task-specific data that could accelerate performance gains in adjacent verticals like pharmaceutical packaging, creating a cross-domain improvement flywheel that competitors without a force-centric simulation core would struggle to replicate.

The size of the win is anchored by the valuation of companies attempting to build general-purpose robot intelligence. A relevant, though not direct, comparable is Figure, which is developing humanoid robots with an emphasis on AI and has reached a reported valuation of $2.6 billion [Business Insider, June 2026]. While Figure is a full-stack hardware and software company, its valuation reflects investor appetite for platforms that promise generalizable robot labor. A pure-play software platform like Eka, if it achieves its goal of becoming the essential intelligence layer for manipulation, could command a significant portion of that value. In a platform-licensing scenario, capturing even a single-digit percentage of the multi-billion dollar industrial robotics software market could support a valuation in the hundreds of millions to low billions within a five-year horizon (scenario, not a forecast).

Data Accuracy: YELLOW -- Core opportunity thesis is inferred from company's stated technical approach and founder backgrounds; specific commercial traction or partnership catalysts are not yet public.

Sources

PUBLIC

1. [ekarobotics.com] Eka Robotics | https://ekarobotics.com/

2. [PitchBook, 2026] Eka Robotics 2026 Company Profile - PitchBook | https://pitchbook.com/profiles/company/1326584-71

3. [LinkedIn] Eka Robotics | https://www.linkedin.com/company/eka-robotics

4. [theresanaiforthat.com] Vision-Force-Action (VFA) | https://theresanaiforthat.com/model/vision-force-action-vfa/

5. [clauday.com] Eka Robotics | https://clauday.com/article/49c4476f-1440-4e89-a0d9-f1f9021353d3

6. [bvp.com, 2026] Robotics that masters physics through self-supervised learning and sim-to-real reinforcement learning. | https://www.bvp.com/atlas/robotics-that-masters-physics-through-self-supervised-learning-and-sim-to-real-reinforcement-learning

7. [csail.mit.edu, 2026] Pulkit Agrawal receives the IROS Toshio Fukuda Young Professional Award | MIT CSAIL | https://www.csail.mit.edu/news/pulkit-agrawal-receives-iros-toshio-fukuda-young-professional-award

8. [professional.mit.edu, 2026] Pulkit Agrawal - MIT Professional Education | https://professional.mit.edu/programs/faculty-profiles/pulkit-agrawal

9. [Business Insider, Jun 2026] Meet the 22 Investors to Know in Robotics and Physical AI | https://www.businessinsider.com/investors-to-know-in-robotics-and-physical-ai-2026-6

10. [TechCrunch, 2024] Skild AI raises $300M Series A | https://techcrunch.com/2024/08/12/skild-ai-raises-300m-series-a/

11. [Figure, 2024] Figure announces $675M Series B | https://www.figure.ai/news/figure-announces-675m-series-b