Walden Robotics

Walden Robotics is an American robotics company based in Cambridge, Massachusetts, that builds wheeled, two-armed general-purpose robots for factories and warehouses. It spun out of the Toyota Research Institute in January 2026 and emerged from stealth on 2026-07-15 with $300 million in seed funding at a stated valuation of $1.1 billion, a round co-led by Toyota and the New York venture firm Deviation Capital.^[1][2][8] The company is led by co-founder and CEO Russ Tedrake, an MIT professor who previously ran the Large Behavior Model program at Toyota Research Institute (TRI), and its technical pitch rests on that lineage: Walden says its robots are driven by large behavior models (LBMs) and diffusion policy, two lines of work its founding team helped originate at TRI.^[1][7] Walden says its robots have been performing production work at a Toyota plant in North America since February 2026, a claim that, as of late July 2026, rests on the company's own account.^[1][7]

The company is named after Walden Pond in Concord, Massachusetts, and Henry David Thoreau's book about it. Walden's stated reason is that it believes "robots will help us find more purpose in our work and in our lives."^[2]

History

Tedrake spent years at TRI as senior vice president for large behavior models, running the group that produced the diffusion policy line of research and, later, TRI's multitask manipulation models.^[1][7] Walden was incorporated as a TRI spinout in January 2026 and operated quietly for about six months before its public launch on 2026-07-15.^[1][7] Toyota did not merely permit the spinout: Toyota Motor Corporation and two Toyota investment arms co-led the financing, and Toyota is also the company's first disclosed deployment site.^[1][7]

The founding team is drawn from TRI, MIT, Stanford, and Amazon.^[1] The Boston Globe named four of the leaders and their prior employers.^[8]

Scroll sideways for more →

Walden has not disclosed its headcount.

Funding

Scroll sideways for more →

Walden and its press materials describe the round as a seed round, which makes it one of the largest first rounds on record in robotics.^[6] The Boston Globe, citing PitchBook, reported that the deal is one of the largest ever for Boston's robotics ecosystem, trailing only fundraising rounds by the autonomous vehicle company Motional.^[8] The $1.1 billion figure is the company's, and no filing confirming it has been published. Deviation Capital founding partner Colin Beirne said in the release that "Walden sits right at the center of that shift" toward more adaptable robots.^[6] Hiroki Nakajima of Toyota said that "Walden's uniqueness is its ability to deliver robots that provide value from day one."^[6]

Technology

Walden describes itself as a full-stack physical AI company, meaning it builds its own hardware, its own software, its own models, and the application layer on top.^[1][6] Tedrake has said the team invented several of the core techniques now used across the embodied AI and humanoid startup field, a claim that is defensible for diffusion policy specifically and harder to assess more broadly.^[7]

Large behavior models

The two research threads Walden names are diffusion policy and large behavior models, which the company says let its robots learn manipulation and decision-making skills instead of following pre-programmed workflows.^[14] Both trace to work Tedrake co-authored.

Diffusion policy, introduced in a 2023 paper by Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Tedrake, and Shuran Song, represents a robot's visuomotor policy as a conditional denoising diffusion process rather than as a direct regression from observation to action. The paper reported an average improvement of 46.9 percent over prior methods across 12 tasks drawn from four manipulation benchmarks, and attributed the gain to receding horizon control, visual conditioning, and a time-series diffusion transformer.^[12]

Large behavior models extend that approach to multitask policies. TRI's 2025 study "A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation" trained diffusion-based LBMs on roughly 1,700 hours of robot data, comprising about 468 hours of in-house bimanual teleoperation, 45 hours collected in simulation, 32 hours of UMI data, and roughly 1,150 hours from the Open X-Embodiment dataset. Evaluation ran to 1,800 real-world rollouts and more than 47,000 simulation rollouts across 29 tasks, on bimanual Franka Panda FR3 arms with up to six cameras. The architecture is a multitask diffusion policy with a multimodal vision transformer encoder and a transformer denoising head conditioned on encoded observations, predicting 16-step action sequences from wrist cameras, scene cameras, proprioception, and a language prompt.^[11][13]

The study's headline results were that pretrained LBMs learned challenging new tasks with three to five times less data than from-scratch baselines, and that performance rose smoothly as pretraining scale and diversity grew, with no observed discontinuity. Its more pointed contribution was methodological: the authors argued that robot learning results are frequently under-powered statistically, and that subtle choices such as data normalization can dominate the architectural changes a paper claims credit for.^[11][13]

The practical distinction from the vision-language-action models associated with Physical Intelligence and its pi-zero family is one of emphasis rather than a hard boundary. Both families condition on language and images and both output action chunks. TRI's LBM line grew out of diffusion policy and puts its weight on the diffusion action head and on carefully measured multitask pretraining, whereas the VLA framing typically starts from a pretrained vision-language model backbone and adapts it to emit actions. Walden has published no model card, parameter count, or benchmark results of its own, and no third-party evaluation of its models existed as of 2026-07-27.

Hardware

Walden's robot has a humanoid upper body with two arms and a head carrying two eye-like sensors, mounted on a wheeled base rather than legs. The Boston Globe described the machines as orange and white.^[8] Tedrake told The Next Web that the wheeled base is easier to certify as safe inside a factory, and that plant workers "aren't ready, and they don't want them yet" when it comes to legged machines.^[9] The Robot Report likewise reported that the form factor was chosen with factory safety certification in mind.^[7]

The company has not announced a product name for the robot, has not published specifications such as height, payload, reach, or battery life, and has not said how many units it has built.

Deployments

Walden says its robots have been doing useful production work at a Toyota plant in North America since February 2026, moving "from first pilot to real work in under two months."^[1][7] The company has not identified the plant, the number of robots, or the contract terms. Bloomberg reported that at least one robot works eight-hour shifts alongside human teams, loading and unloading car parts, cleaning machinery, and kitting parts for assembly.^[15] Walden's own materials list machine tending, tool setting, parts kitting, and assembly as the target task set.^[3][7]

These are the company's assertions relayed by reporters rather than independently audited claims, and no journalist has published an on-site verification of the deployment. That distinction matters in this sector, where deployment announcements routinely outrun what is running unattended in production. Walden's stated target industries are automotive, aerospace, semiconductors, electronics, logistics, and life sciences, and it has named no customer other than Toyota.^[1][5]

Business model

Walden does not sell robots outright. The Boston Globe reported that the company charges usage fees under a robots-as-a-service arrangement, a choice Walden justified by how quickly the underlying technology is changing.^[8] Walden's website carries a "hire a robot" call to action rather than a purchase path.^[3] This places it alongside the robotics-as-a-service operators rather than the equipment vendors.

The company is also explicit that its robots are not fully autonomous in deployment. Its own launch commentary describes the offering as "powerful autonomy combined with a human available for remote assistance, providing a fully-capable robot immediately upon deployment."^[2] In other words, a remote human operator covers the gap between what the models can do and what the customer needs done, which is also how the company keeps collecting the real-world data its models train on. Walden has not said what fraction of task time is autonomous, a figure that is the central open question for any company making this claim. See humanoid robot autonomy levels for the general framework.

Competitive position

Walden entered a crowded field. Figure AI, Apptronik, 1X Technologies, Agility Robotics, and Tesla Optimus are all pursuing general-purpose robots for industrial or domestic work, and Skild AI and Physical Intelligence are building the models without the hardware. Walden differs from most of them on two axes at once.

The first is the body. Most of the best-funded entrants build bipeds, and the humanoid robot label carries the legged form with it. Walden chose wheels, which narrows where the robot can go but simplifies safety certification, balance, and battery budget. That puts it closer to wheeled-base manipulators such as the Reflex Robotics humanoid than to Figure's or Tesla's machines. It also echoes Toyota's own earlier wheeled manipulator, the Human Support Robot.

The second is the go-to-market. Selling robot-hours rather than robots, with a remote operator behind the autonomy, front-loads revenue but caps gross margin until autonomy improves. Tedrake has publicly stressed unit economics as the constraint that decides which of these companies survive.^[9]

Walden's structural advantage is Toyota. A spinout with its former parent as co-lead investor and first production customer gets a real factory to iterate in, which is the scarcest input in this field. The corresponding risk is concentration: one disclosed customer, who is also an investor, is not yet evidence of a market. Broader context on the sector is at humanoid robot manufacturers and humanoid robot deployments.

Reception

Coverage of the launch was heavy and largely descriptive. Trade and business outlets framed the story around the size of the round, the Toyota relationship, and Tedrake's research reputation.^{[7][8][9][10]} The Next Web's angle was the absence of legs, treating Walden's wheeled base as a deliberate contrast with the biped-heavy field.^[9]

Tedrake himself was the most cautious voice in the coverage, telling The Next Web that "success is not assured" and emphasizing unit economics and business planning over technical demonstrations.^[9]

No independent technical evaluation of Walden's models or robots had been published as of 2026-07-27, and the company has released no benchmark results, no video of extended unattended operation, and no autonomy metrics. Every performance and deployment claim in circulation originates with Walden. Given that the company is three weeks old in public and that its valuation exceeds a billion dollars on the strength of one undisclosed deployment, the gap between what has been asserted and what has been verified is the appropriate frame for reading it.

See also

• Toyota Research Institute
• Large Behavior Model
• Diffusion policy
• Robot foundation model
• Robotics as a Service
• Humanoid robot manufacturers
• Humanoid robot market

References

1. "Walden Robotics Launches with $300 Million to Put General-Purpose Robots to Work Today." Walden Robotics, 2026-07-15. https://www.waldenrobotics.com/news/walden-robotics-launches-from-stealth ↩
2. "Why Walden?" Walden Robotics, 2026-07. https://www.waldenrobotics.com/news/why-walden ↩
3. "Build the Future of Physical AI with Us." Walden Robotics, accessed 2026-07-27. https://www.waldenrobotics.com/ ↩
4. "Walden Robotics Launches with $300 Million to Put General-Purpose Robots to Work Today." Business Wire, 2026-07-15. https://www.businesswire.com/news/home/20260715089377/en/ ↩
5. "Walden Robotics Launches with $300 Million to Put General-Purpose Robots to Work Today." Yahoo Finance, 2026-07-15. https://finance.yahoo.com/technology/ai/articles/walden-robotics-launches-300-million-100000118.html ↩
6. "Walden Robotics Launches with $300 Million to Put General-Purpose Robots to Work Today." RoboticsTomorrow, 2026-07-15. https://www.roboticstomorrow.com/news/2026/07/15/walden-robotics-launches-with-300-million-to-put-general-purpose-robots-to-work-today/26842/ ↩
7. "Walden Robotics launches at $1.1B valuation for general-purpose robots." The Robot Report, 2026-07-15. https://www.therobotreport.com/walden-robotics-launches-1-1b-valuation-general-purpose-robots/ ↩
8. "Stealthy Toyota spinout gets one of Boston area's biggest robotics deals ever." The Boston Globe, 2026-07-15. https://www.bostonglobe.com/2026/07/15/business/walden-robotics-cambridge-toyota-humanoid/ ↩
9. "Walden Robotics launches with $300M, and no legs." The Next Web, 2026-07-15. https://thenextweb.com/news/walden-robotics-300-million-toyota-humanoid ↩
10. "Walden Robotics launches humanoid robots for manufacturing and logistics." DC Velocity, 2026-07. https://www.dcvelocity.com/material-handling/robotics/walden-robotics-launches-humanoid-robots-for-manufacturing-and-logistics ↩
11. TRI LBM Team. "A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation." arXiv:2507.05331, 2025-07-07. https://arxiv.org/abs/2507.05331 ↩
12. Chi, Cheng, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. "Diffusion Policy: Visuomotor Policy Learning via Action Diffusion." arXiv:2303.04137, 2023-03-07. https://arxiv.org/abs/2303.04137 ↩
13. "A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation" (project page). Toyota Research Institute, 2025. https://toyotaresearchinstitute.github.io/lbm1/ ↩
14. "Video: US-based Toyota spinout's factory robots learn from experience on the job." Interesting Engineering, 2026-07. https://interestingengineering.com/ai-robotics/us-toyota-robots-learn-from-experience ↩
15. "Toyota-Backed Startup Walden Robotics Comes Out of Stealth With $1.1 Billion Valuation." Bloomberg, 2026-07-15. https://www.bloomberg.com/news/articles/2026-07-15/toyota-backed-robotics-startup-walden-launches-with-1-1-billion-valuation ↩