# Rhoda AI

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

**Rhoda AI** (also called **Rhoda**) is an American robotics company that builds [robot foundation models](https://aiwiki.ai/wiki/robot_foundation_model) pre-trained mainly on internet video, along with its own robot hardware for factory and warehouse work. It was founded in 2024 by Jagdeep Singh, the co-founder and former chief executive of the energy-storage company QuantumScape, together with [Stanford University](https://aiwiki.ai/wiki/stanford_university) professor Gordon Wetzstein, Eric Ryan Chan, Changan Chen and Andrew Wooten.[7][9][15][16] The company calls its control approach a Direct Video-Action (DVA) model: a causal video model predicts the next moments of the robot's camera view, and a separate inverse dynamics model turns that predicted video into motor commands, in a closed loop.[4] Rhoda markets the resulting system as FutureVision.[1]

Rhoda left stealth on March 10, 2026, after what it said were 18 months of private development, announcing a $450 million Series A round.[1] Bloomberg reported that the round was led by Premji Invest and valued the company at $1.7 billion including the new money; Reuters also reported the $1.7 billion figure.[2][3] The company's own announcement did not state a valuation.[1] Since the launch Rhoda has published research on the data efficiency of its models, on scaling video pre-training, and on a tactile sensor of its own design, and its chief executive has described a wheeled mobile manipulator with seven-degree-of-freedom arms that the company is developing in house.[4][5][6][11]

## Overview

| Field | Detail |
| --- | --- |
| Founded | 2024; public launch March 10, 2026[1][16] |
| Co-founders | Jagdeep Singh (CEO), Eric Ryan Chan (Chief Scientist), Gordon Wetzstein (Scientific Advisor), Andrew Wooten (Chief Product Officer), Changan Chen (Chief Research Officer)[7] |
| Headquarters | Palo Alto, California (launch announcement dateline); most 2026 job listings were for Mountain View, California, with some in Munich, Germany[1][27] |
| Focus | Video-pretrained robot control models and full-stack industrial robots[1][12] |
| Core technology | Direct Video-Action (DVA) model, branded FutureVision[1][4] |
| Latest disclosed round | $450 million Series A, led by Premji Invest (March 2026)[1][2] |
| Reported valuation | $1.7 billion including the money raised (Bloomberg, March 2026)[2] |
| Customer sectors (company statement) | Automotive, manufacturing, logistics and e-commerce[8] |
| Website | rhoda.ai[8] |

## History

### Background and founding

Singh founded the telecommunications company Infinera and was its chief executive from 2001 to 2009. He co-founded QuantumScape in 2010 and was its chief executive until February 15, 2024, and its chairman until he retired from the board on December 31, 2024.[13][14][15] In an October 2026 podcast interview, he said he had come to robotics from outside the field and concluded that its central problem was a shortage of training data, because the dominant approach collected robot data through [teleoperation](https://aiwiki.ai/wiki/teleoperation).[12] He said the idea of learning from internet video came out of conversations with Wetzstein, with whom he had previously started the display company Raxium, which they sold to Google, and that Wetzstein then pointed him to Eric Chan, one of Wetzstein's PhD students.[12] Wetzstein's Stanford curriculum vitae lists him as a co-founder of Raxium from November 2017 to May 2022 ("acquired by Google") and as a co-founder of Rhoda AI since October 2024.[16] Google confirmed in May 2022 that it had acquired Raxium, a developer of microLED displays.[30]

While Rhoda was still in stealth, Forbes reported in October 2025 that it had raised a $162.6 million Series A round in April 2025, bringing its total funding to $230 million and valuing it at nearly $1 billion, figures the magazine attributed to PitchBook.[9] Forbes also reported, citing documents it had seen, that the company was developing a "general purpose bimanual manipulation platform" and that its founding team included Singh, Wetzstein and Vincent Clerc, who had worked on SoftBank's [Pepper](https://aiwiki.ai/wiki/pepper_robot) robot; Rhoda declined to comment.[9] Rhoda has not publicly explained how that earlier round relates to the $450 million Series A it announced in March 2026, and Clerc was not among the people listed on the company's team page in October 2026.[7]

### Public launch (March 2026)

On March 10, 2026, Rhoda announced its public launch after 18 months in stealth, unveiled FutureVision and said it had raised $450 million in Series A funding.[1] The announcement named Capricorn Investment Group, [Khosla Ventures](https://aiwiki.ai/wiki/khosla_ventures), Leitmotif, Matter Venture Partners, Mayfield, Premji Invest, Prelude Ventures, Temasek and Xora as backers, along with John Doerr.[1] Bloomberg's Dina Bass reported the same day that Premji Invest led the round and that it valued Rhoda at $1.7 billion "including the money raised."[2] Reuters reported that the Series A valued the company at $1.7 billion and that Rhoda expected eventually to license FutureVision to companies running robotic hardware and software platforms.[3]

The company said the money would go to research and engineering, more industrial deployments and customer pilots, and hiring across generative AI, computer vision and robotics.[1] It also said that in "a recent high-volume manufacturing evaluation" its system had completed a component-processing workflow in under two minutes per cycle without human intervention, exceeding the customer's targets; it did not name the customer.[1] SiliconANGLE, drawing partly on Singh's comments to Bloomberg, reported that the company had run successful tests with an automotive firm using off-the-shelf parts and that Singh said Rhoda planned to build humanoid-style robots and make its own hardware.[22] Investing News Network similarly reported tests on off-the-shelf robotic hardware inside an automotive manufacturing facility.[25]

### Research releases and hardware plans (2026)

Rhoda published a long technical post describing the DVA approach on the day of its launch, a study of how scaling video pre-training affects robot performance on September 10, 2026, and a post on tactile sensing on October 13, 2026.[4][5][6] In interviews with Brian Heater of the Association for Advancing Automation (A3) in September and October 2026, Singh said the company had been testing its own hardware with customers but had not yet announced it, and that Rhoda planned to ship its model on off-the-shelf hardware first.[11][12]

A job listing posted in June 2026 described a European subsidiary, Rhoda AI Europe GmbH, as the company's "first international entity," set up to manage partner relationships in Europe and to deploy robots under a robotics-as-a-service model. The same listing said the company would have "the first industrial deployments in the field by Q4-2026."[26] As of October 2026 the company's team page listed 62 people.[7]

## Technology

### Direct Video-Action model

Rhoda argues that [vision-language-action models](https://aiwiki.ai/wiki/vision_language_action_model), which start from a pre-trained vision-language model and post-train it on robot teleoperation data, collect tens or hundreds of thousands of hours of robot data yet remain far from general behavior, and that web video is a far larger source of information about how the physical world moves.[4][12] Its DVA model treats robot control as real-time video prediction. Conditioned on a long history of the robot's camera frames, proprioception and optional language instructions, a causal video model generates a short prediction of what should happen next; an inverse dynamics model then converts the predicted frames into end-effector motions, and the cycle repeats multiple times per second.[4] The company's launch announcement described the loop as running "every few hundred milliseconds."[1]

The model is pre-trained on general web video, "hundreds of millions of videos" according to the company, with a causal generation objective, and then post-trained on robot video for particular tasks and robot types.[1][4] Rhoda says it trains its video model as a causal model from scratch rather than distilling one from a bidirectional video model, and states that, to the best of its knowledge, its model is "the first to pre-train a causal video model from scratch and also the first to perform full video denoising during real-time closed-loop robot control."[4] Its write-up cites earlier work that used video models to synthesize robot training data (such as GR00T N1 and UniSim), fine-tuned video models to predict actions (such as GR-2 and Cosmos Policy), planned open-loop with video models, or ran closed-loop control with video models distilled from non-causal ones.[4] Other techniques it describes include:

| Technique | Description (as described by Rhoda) |
| --- | --- |
| Context Amortization | During training, the model predicts future frames at every position in a long, noise-free video history, much as a language model predicts the next token at every position, so that hundreds of frames of context can be used efficiently. Rhoda contrasts this with [Diffusion Forcing](https://aiwiki.ai/wiki/diffusion_forcing), in which each frame is noised independently.[4] |
| Inverse dynamics model | A small, separate model that translates predicted video into actions. Rhoda says it can be trained on about 10 hours of data per robot type, including random motions rather than task demonstrations, and reused across tasks.[4] |
| Leapfrog Inference | Each prediction looks far enough ahead to cover the latency of the next one, and each new prediction is conditioned on the action currently being executed, to keep motion continuous.[4] |
| Long-context memory | The model keeps hundreds of frames of visual context, which Rhoda says lets it track progress through multi-step tasks without hand-built task-progress scaffolding.[4] |
| Interpretability through video | Because the policy produces video first, engineers can generate rollouts to inspect, debug and compare models.[4] |

At inference time the model reuses its encoded context through [KV caching](https://aiwiki.ai/wiki/kv_cache) and always conditions on real camera observations rather than its own earlier generations.[4]

### Demonstrated tasks

Rhoda's March 2026 post presented two customer tasks that it said had been deployed as proofs of concept and had run for hours without human intervention.[4] In a bearing-decanting task from an automotive assembly line, a robot uses both arms to unpack 10 kg boxes of bearings into a tote and sort the packaging; the company said it used 11 hours of robot data and showed an uncut 1.5-hour autonomous run.[4][8] In a container-breakdown task, the robot unlatches and collapses Contico containers of about 50 pounds and clears debris from them; Rhoda said it used 17 hours of robot data and showed a 160-minute continuous run.[4] The post also showed an end-to-end returns-processing task (unpacking, inspecting, folding, repackaging and binning clothes) for a logistics customer, a "shell game" that tests visual memory, and one-shot imitation of a single human demonstration for sorting objects and for drawing shapes in the demonstrated stroke order.[4][8]

Singh has said the container-folding model, first trained in Rhoda's own facility, struggled in the customer's factory because the containers there were worn, some latches were broken and rusted transfer tables changed the friction, and that about three hours of additional training data collected in the factory was enough for it to run autonomously.[10][12] He has also said that pre-training worked better on broad, less curated video than on footage of workers doing the target job, because the purpose of pre-training is to learn general physical dynamics rather than any specific task.[10][12]

### Scaling study (September 2026)

On September 10, 2026, Rhoda published a study of whether scaling pre-training on general web video improves performance on a real industrial task. It pre-trained four model sizes (labeled XS, S, M and L) and, separately, the M model at four compute budgets, then post-trained each on the same bearing-unpacking demonstrations and evaluated them on a real robot.[5] The metric was the "at-speed completion rate," the share of trials that achieved every goal of the task without human intervention in under 100 seconds, a limit the company said was stricter than its customer's requirement.[5] The study took over 200 hours of real-robot evaluation, with one hundred to a few hundred trials per policy.[5]

| Experiment | Reported at-speed completion rate |
| --- | --- |
| Model size XS, S, M, L (same recipe) | 3.7%, 65.0%, 75.3%, 84.7%[5] |
| M model at 0.08x, 0.18x, 0.37x, 1x of its pre-training compute | 57.8%, 67.3%, 74.5%, 75.3%[5] |

Rhoda reported that larger models were consistently better, that more pre-training compute helped most when robot data was scarce, and that a model's ability to predict held-out web video, measured as a Fréchet distance between [DINOv2](https://aiwiki.ai/wiki/dinov2) features of predicted and real frames, correlated with its robot performance across all seven checkpoints.[5] The company listed limitations: the study covered one task, one robot embodiment and one setup, and the relationship between pre-training quality and task performance rested on seven checkpoints of one architecture family.[5] The report cited published robot scaling results from [Generalist AI](https://aiwiki.ai/wiki/generalist_ai) (GEN-0), [Dyna Robotics](https://aiwiki.ai/wiki/dyna_robotics) (Dyna-2) and NVIDIA (EgoScale), noting that those used manipulation data with action labels, whereas Rhoda's pre-training data was general web video with no actions.[5]

### Tactile sensing (October 2026)

On October 13, 2026, Rhoda described a [tactile sensor](https://aiwiki.ai/wiki/tactile_sensing) it had developed and a multimodal version of DVA that takes a history of tactile signals as input and generates future tactile signals, which the inverse dynamics model turns into actions.[6] The company reported that the sensor responded to a feather's touch of a few millinewtons and to loads of over 40 newtons, distinguished normal from shear forces, sampled contact vibrations at about 500 Hz and registered temperature changes; it said off-the-shelf optical and magnetic sensors it tested showed no distinguishable response to the feather.[6] Rhoda also fitted the sensors to a handheld [Universal Manipulation Interface](https://aiwiki.ai/wiki/universal_manipulation_interface) to record contact forces during human demonstrations.[6] In a demonstration with the robot's wrist and head cameras blocked, the touch-guided model searched a box, identified a textured ball by touch, grasped it and placed it beside the box, and the company said the same model generalized to a box of seven balls including textured balls of different size, firmness and texture.[6]

## Hardware

Rhoda's website says its robot platform is designed with custom actuators ("25kg rated, 40kg peak"), a wheeled base, brakes in every actuator, safety-rated vision and "3 years of continuous operation at rated payload."[8] Singh has said the company intended to be full-stack from the start and builds its own hardware because the third-party robots it tried fell short on payload, reliability, safety and dexterity; he said that after buying a few humanoids, Rhoda burned out one motor within a day and a second within a week.[12] He described the robot as a mobile manipulator rather than a legged humanoid, with seven-degree-of-freedom arms, dexterous hands and a vision system on a wheeled base, and said that with a wheeled base the robot does not collapse when it is emergency-stopped.[11][12] He said Rhoda wants its robot to lift about twice the OSHA guideline for people in the United States, or about 45 kg close to the body; wants every subsystem to be CE certified or certifiable; is building its own actuators and tactile sensors; and considers a three-fingered hand sufficient for most tasks, though it was still weighing the options.[11][12] He also said there were "probably over 100 humanoid companies around the world, but most of them, to be candid, are toys," because they could lift little weight and could not do so reliably day after day.[11][12] Rhoda's 2026 job listings referred to "humanoid robots" and "humanoid platforms" and included roles for robot pilots who collect manipulation data by VR teleoperation.[27]

## Business model

Rhoda says FutureVision powers its own systems and is expected over time to be licensed to partners on other robot hardware and software platforms.[1] Singh has described a single model that is pre-trained on video and post-trained on customer tasks, with the latest checkpoint post-trained further on each new customer's data; he said Rhoda requires customers to grant it an irrevocable, perpetual license to train on data collected in their facilities, while the customers keep ownership of the data.[12] He has said the company's model typically needs 10 to 20 hours of task data to start, with smaller amounts of edge-case data collected in production used to improve it further.[12] The company says it works with customers in automotive, manufacturing, logistics and e-commerce.[8] Jens Wiese, a managing partner at investor Leitmotif and a former Volkswagen Group executive, said in the launch announcement that technology like Rhoda's "can dramatically expand the scope of what can be automated."[1]

## Funding

| Date | Round | Amount | Valuation | Investors and notes |
| --- | --- | --- | --- | --- |
| April 2025 (reported October 2025) | Series A, as described by Forbes citing PitchBook | $162.6 million; $230 million raised in total | Nearly $1 billion (PitchBook figure reported by Forbes) | Rhoda declined to comment[9] |
| March 10, 2026 | Series A (announced by the company) | $450 million | $1.7 billion including the money raised, according to Bloomberg; $1.7 billion according to Reuters | Led by Premji Invest; also Capricorn Investment Group, Khosla Ventures, Leitmotif, Matter Venture Partners, Mayfield, Prelude Ventures, Temasek, Xora and John Doerr[1][2][3] |

Company job listings in 2026 said Rhoda had "raised over $450M."[26] As of October 2026 the investor section of Rhoda's website also displayed logos of firms not named in the March announcement, among them [NVentures](https://aiwiki.ai/wiki/nvidia_nventures), Samsung Next, Gigascale Capital, Hyperlink Ventures, StepStone, MVP Ventures, ROC Venture Group, Knollwood, Strike and B37; the site does not say when or in which round they invested.[8]

## Leadership

| Person | Role at Rhoda | Background |
| --- | --- | --- |
| Jagdeep Singh | Co-founder and CEO[7] | Founder and CEO of Infinera (2001-2009); co-founder of QuantumScape, its CEO (2010-2024) and chairman until December 2024; B.S. in computer science from the University of Maryland, M.S. in computer science from Stanford, MBA from UC Berkeley's Haas School of Business[14][15] |
| Eric Ryan Chan | Co-founder and Chief Scientist[7][17] | Listed among Wetzstein's former PhD students (Stanford computer science, 2021-2024; NVIDIA Graduate Fellowship);[16] first author of the 3D-aware [generative adversarial network](https://aiwiki.ai/wiki/generative_adversarial_network) papers pi-GAN and EG3D;[18][19] previously a generative model architect at WorldLabs, according to Rhoda's launch announcement[1] |
| Gordon Wetzstein | Co-founder and Scientific Advisor[7] | Associate professor of electrical engineering (and, by courtesy, computer science) at Stanford; co-founder of Raxium (acquired by Google) and of Zinn Labs[16] |
| Changan Chen | Co-founder and Chief Research Officer[7] | Former Stanford postdoctoral researcher hosted by [Fei-Fei Li](https://aiwiki.ai/wiki/fei_fei_li) and Ehsan Adeli; PhD from the University of Texas at Austin advised by Kristen Grauman; visiting researcher at Facebook AI Research during his PhD[20] |
| Andrew Wooten | Co-founder and Chief Product Officer[7] | Leads product strategy and finance; formerly chief revenue officer at Crimson Education; Stanford MBA, according to a conference speaker biography[21] |
| Steve Tirado | Chief Strategy Officer[7] | |
| Alex Bergman | Chief Data Officer and VP of Software Engineering[7] | |

Rhoda's launch announcement described Wetzstein as head of Stanford's Computational Imaging Lab; by October 2026 his faculty page listed him as director of the Stanford Physical and Spatial Intelligence Lab.[1][16]

## Reception

Launch coverage focused on Rhoda's use of internet video in place of large teleoperated robot datasets.[2][22][23][24] Reuters placed the launch in a race over humanoid robots led by [Tesla](https://aiwiki.ai/wiki/tesla), [Figure AI](https://aiwiki.ai/wiki/figure_ai), [Unitree](https://aiwiki.ai/wiki/unitree), [Agility Robotics](https://aiwiki.ai/wiki/agility_robotics) and dozens of Chinese startups, and noted that industry experts saw reliability, safety certification and cost as the main hurdles to large-scale deployment of general-purpose robots.[3] Forbes's October 2025 report presented Rhoda and [Genesis AI](https://aiwiki.ai/wiki/genesis_ai) as two Silicon Valley startups that had each raised more than $100 million while secretly developing humanoid robots.[9]

The robotics news site Humanoids Daily wrote in its company profile that public evidence of deployment was "limited and mostly company-supplied."[28] After Singh told him in September 2026 that Rhoda could be the first autonomous manipulation-capable robot deployed at scale, A3's Brian Heater wrote that beating the many industrial humanoid makers to market "would certainly be a coup for a system that has yet to be announced."[11] Singh himself has said that despite billions of dollars raised across the sector he could not think of "a single example of a generalist autonomous robot that's doing useful work in the field today."[11][12]

Debarghya Das's "Ultimate Neolab List" includes Rhoda and rates it "Mostly" a neolab, with the note: "Robot-intelligence model company with a strong research team, but it deploys and plans its own hardware."[29]

## See also

- [List of neolabs](https://aiwiki.ai/wiki/list_of_neolabs)
- [Neolab](https://aiwiki.ai/wiki/neolab)
- [Robot foundation model](https://aiwiki.ai/wiki/robot_foundation_model)
- [Vision-language-action model](https://aiwiki.ai/wiki/vision_language_action_model)
- [World model](https://aiwiki.ai/wiki/world_model)
- [Physical Intelligence](https://aiwiki.ai/wiki/physical_intelligence)
- [Genesis AI](https://aiwiki.ai/wiki/genesis_ai)
- [Physical AI](https://aiwiki.ai/wiki/physical_ai)

## References

1. Rhoda AI, [Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World](https://rhoda.ai/news/press-release) (press release), March 10, 2026.
2. Dina Bass, Bloomberg, [AI Robotics Startup Rhoda Valued at $1.7 Billion in New Funding](https://www.bloomberg.com/news/articles/2026-03-10/ai-robotics-startup-rhoda-valued-at-1-7-billion-in-new-funding), March 10, 2026 ([archived copy](http://web.archive.org/web/20260310154521/https://www.bloomberg.com/news/articles/2026-03-10/ai-robotics-startup-rhoda-valued-at-1-7-billion-in-new-funding)).
3. Akash Sriram, Reuters (via Yahoo Finance), [Rhoda AI raises $450 million at $1.7 billion valuation, unveils robot intelligence platform](https://finance.yahoo.com/news/rhoda-ai-raises-450-million-160945418.html), March 10, 2026.
4. Rhoda AI Team, [Causal Video Models Are Data-Efficient Robot Policy Learners](https://rhoda.ai/research/direct-video-action), Rhoda AI Research blog, March 10, 2026.
5. Rhoda AI Research, [Does Scaling Web-Video Pre-training Help Real Robots Do Real Work?](https://rhoda.ai/research/scaling-web-video-pretraining), September 10, 2026.
6. Rhoda AI Research, [Giving Robots a Sense of Touch](https://rhoda.ai/research/giving-robots-a-sense-of-touch), October 13, 2026.
7. Rhoda AI, [Team](https://rhoda.ai/team), accessed October 27, 2026.
8. Rhoda AI, [Home page](https://rhoda.ai/), accessed October 27, 2026.
9. Anna Tong, Forbes, [These Well-Funded AI Startups Are Building Humanoid Robots In Stealth](https://www.forbes.com/sites/annatong/2025/10/15/two-ai-startups-have-each-raised-100-million-to-build-humanoid-robots-in-stealth/), October 15, 2025 ([archived copy](http://web.archive.org/web/20260802142357/https://www.forbes.com/sites/annatong/2025/10/15/two-ai-startups-have-each-raised-100-million-to-build-humanoid-robots-in-stealth/)).
10. Brian Heater, Association for Advancing Automation, [Rhoda AI CEO on Video Training: I Don't Think the World is Going Back](https://www.automate.org/ai/industry-insights/rhoda-ai-makes-a-big-bet-on-video), April 2, 2026.
11. Brian Heater, Association for Advancing Automation, [Rhoda AI's Deliberate Approach to Hardware](https://www.automate.org/ai/industry-insights/rhoda-ai), September 3, 2026.
12. Brian Heater, Automated (A3 podcast), [Jagdeep Singh on Rhoda AI, Internet Video and Robot Learning](https://www.automate.org/automated-podcast/episodes/automated-podcast-episode-jagdeep-singh), episode 63 (with transcript), October 14, 2026.
13. QuantumScape Corporation, [Form 8-K current report](https://www.sec.gov/Archives/edgar/data/1811414/000095017024015280/qs-20240208.htm), U.S. Securities and Exchange Commission, February 8, 2024.
14. QuantumScape Corporation, [Form 8-K current report](https://www.sec.gov/Archives/edgar/data/1811414/000095017024116716/qs-20241017.htm), U.S. Securities and Exchange Commission, October 17, 2024.
15. QuantumScape Corporation, [2024 proxy statement (Schedule 14A)](https://www.sec.gov/Archives/edgar/data/1811414/000114036124022958/ny20019238x2_def14a.htm), U.S. Securities and Exchange Commission, 2024.
16. Gordon Wetzstein, [Faculty home page and CV](https://stanford.edu/~gordonwz/), Stanford University, accessed October 27, 2026.
17. Eric Ryan Chan, [About](https://ericryanchan.github.io/about), personal website, accessed October 27, 2026.
18. Eric R. Chan, Connor Z. Lin, Matthew A. Chan, et al., [Efficient Geometry-aware 3D Generative Adversarial Networks](https://arxiv.org/abs/2112.07945), arXiv:2112.07945, December 2021.
19. Eric R. Chan, Marco Monteiro, Petr Kellnhofer, Jiajun Wu and Gordon Wetzstein, [pi-GAN: Periodic Implicit Generative Adversarial Networks for 3D-Aware Image Synthesis](https://arxiv.org/abs/2012.00926), arXiv:2012.00926, December 2020.
20. Changan Chen, [Personal website](https://changan.io), accessed October 27, 2026.
21. RAISE Summit, [Andrew Wooten speaker profile](https://tickets.raisesummit.com/Machina/speaker/2354099/andrew-wooten), accessed October 27, 2026.
22. Kyt Dotson, SiliconANGLE, [Rhoda AI raises $450M to build foundational robotics models that learn from internet videos](https://siliconangle.com/2026/03/10/rhoda-ai-raises-450m-build-foundational-robotics-models-learn-internet-videos/), March 10, 2026.
23. Robotics 24/7 Staff, [Rhoda AI exits stealth mode with $450M Series A](https://www.robotics247.com/article/rhoda-ai-exits-stealth-mode-with-450m-series-a), March 10, 2026.
24. The Robot Report Staff, [Rhoda AI exits stealth with $450M to train robots from video](https://www.therobotreport.com/rhoda-ai-exits-stealth-with-450m-to-train-robots-from-video/), March 12, 2026.
25. Giann Liguid, Investing News Network, [AI Robotics Startup Rhoda Hits US$1.7 Billion Valuation After Successful Funding Round](https://investingnews.com/rhoda-valuation-surpasses-one-billion/), March 11, 2026.
26. Rhoda AI, [Field Deployment Lead (job listing, Munich)](https://jobs.ashbyhq.com/rhoda-ai/c20aae07-aaf9-4ed8-bbae-f9888d5b02c5), posted June 22, 2026, accessed October 27, 2026.
27. Rhoda AI, [Rhoda AI jobs](https://jobs.ashbyhq.com/rhoda-ai), Ashby job board, accessed October 27, 2026.
28. Humanoids Daily, [Rhoda AI company profile](https://www.humanoidsdaily.com/companies/rhoda-ai), accessed October 27, 2026.
29. Debarghya Das, [The Ultimate Neolab List](https://debarghyadas.com/neolabs/), accessed October 27, 2026.
30. Duncan Riley, SiliconANGLE, [Google confirms it has acquired microLED startup Raxium](https://siliconangle.com/2022/05/04/google-confirms-acquired-microled-startup-raxium/), May 2022.