Dyna Robotics Taku: The Wheeled Robot That Folded Laundry for an Hour on Its Own

Dyna Robotics says its new wheeled semi-humanoid, Taku, ran an uncut, one-hour hotel laundry-room workflow on its own, according to the company's 29 September 2026 research post. Taku has two 7-degree-of-freedom arms, a folding lower body and four steerable wheels. Dyna has published no price, weight or battery figures, and the one-hour claim comes with no measured success rate.

What is Dyna Robotics' Taku?

Taku is the physical body of Dyna-2.1, which Dyna calls a "full-stack physical agent" that can complete end-to-end real-world workflows. Dyna describes it as having "a human-shaped body above the waist, a folding lower body to reach low and high, an incredibly stable base on four steerable wheels, and two 7-degree-of-freedom arms," and says it sized Taku "like an average person."

In plain terms: it is a torso with two arms on a wheeled base that can crouch and stretch, built to do shift work in rooms designed for people. It does not walk.

What did the one-hour laundry demo actually show?

Dyna says an uncut one-hour video shows Dyna-2.1 running a hotel laundry-room workflow autonomously. According to the launch press release, the tasks included loading the washer and dryer and turning them on, unloading bulk towels from the back of a dryer, folding towels and stacking them by size, placing stacks on bottom and top shelves, and recovering from errors such as a dropped towel.

All of this is the company's own account. Dyna does not name the site, and it publishes no success rate, no count of human interventions and no figure for Mean Time Between Interventions (MTBI), the metric its press release calls a more useful benchmark than per-episode success rates. Dyna also says that, to the best of its knowledge, Dyna-2.1 is the first physical agent to perform hour-long, non-linear loco-dexterous manipulation workflows. No independent party has confirmed that.

What are Taku's specifications?

Dyna has published a short list of hardware facts and left the rest blank. Every entry below is company-stated; none has been independently measured. Figures are as of 12 October 2026.

Dyna says the planetary actuators allow higher speeds and accelerations than the harmonic drives used in conventional arms. That is a design claim with no test data attached.

How does Taku work?

Taku runs three layers of software, per Dyna. A vision-language model acts as a workflow orchestrator: it tracks where the job stands, decides the next step and steers the policy. The Dyna-2 policy, which Dyna calls an improved version of its world-action model, runs at about 5 Hz. A whole-body controller trained with reinforcement learning in simulation then outputs joint targets and wheel velocities at 100 Hz, and the page shows that training in NVIDIA Isaac Lab.

The split matters for a normal user. The slow layer decides what to do next, the fast layer keeps the wheels and arms coordinated, so the robot can drive, bend and reach as one motion rather than in separate steps.

On training data, Dyna says it pre-trains the action model on "one million hours of human video mixed with robot data from our fleet." The company also says Dyna-2, pre-trained on human video alone, passed 87% of customer acceptance tests zero-shot on a stationary robot, versus 46% for Dyna-1. That figure is lab-reported and refers to a stationary robot, not to Taku.

What is Dyna's thesis, and how is it funded?

Dyna's bet is that a robot is worth paying for only when it finishes a whole workflow without a person nearby. CEO Lindon Gao said in the press release: "Our customers need robots that complete an entire workflow for a full shift, without human babysitters."

The Letter Two names the founders as Lindon Gao, York Yang and former DeepMind research scientist Jason Ma. The press release says Gao and Yang sold Caper AI for $350 million.

On funding, The Robot Report says Dyna closed a $120 million Series A led by RoboStrategy, CRV and First Round Capital. Salesforce Ventures, NVentures, the Amazon Industrial Innovation Fund, Samsung Next and LG Technology Ventures participated. The Letter Two reports nearly $144 million raised in total and says the company had not disclosed its valuation. No later round turned up as of 12 October 2026.

How does Taku compare with other wheeled robots?

Taku sits among wheeled, two-armed platforms such as Galbot's G1. Unlike the Astribot T1, which is sold as a developer platform with a public price, Taku has no listed price or full spec sheet. Figures are company-stated as of 12 October 2026, and gaps are marked.

The technology differences are mostly about intent. Astribot, per its press release, emphasises a cable-driven body and says the T1 does not learn autonomously during deployment. Dyna emphasises a hierarchy of orchestrator, policy and whole-body controller aimed at long jobs. Galbot is a Beijing company whose G1 is designed to automate inventory, replenishment, delivery and packaging, and which raised about RMB 1.1 billion (around $153 million) in a round reported on 2 July 2025.

Where will people actually use Taku?

Dyna says Dyna-2.1 is being deployed in hotels, laundromats and restaurants, and names food preparation, data center management and commercial laundry as workflows. Realistic uses, in plain terms:

1. Hotel laundry rooms. Load machines, move loads, fold and shelve towels.
2. Laundromats. The same loop on a larger scale and with more varied items.
3. Restaurant back rooms. Food preparation is one of the named workflows.
4. Shelf work at mixed heights. The folding lower body is designed to reach low and high shelves.

What's still unclear?

The weakest point is evidence. A single uncut video proves the robot can do the job once, not how often it needs help. Dyna's press release champions MTBI as the better benchmark but gives no number, so buyers cannot yet compare it with a human worker or another robot.

Customers are also unclear. The press release says the system is being deployed in hotels, laundromats and restaurants and names no customer. Dyna's research post names none either, and describes taking the new system to real-world customer sites as its next milestone. Interesting Engineering reports the deployment claim as the company's. Whether Taku itself, as opposed to Dyna's earlier stationary systems, is working at a paying site is not established.

Hardware basics are missing: weight, payload, compute, battery and price. Dyna does say a new site can reach its production bar in as little as three days, but that figure is for stationary folding deployments, not Taku.

Should you buy it or wait?

You cannot buy it as a consumer, so the question is for businesses. Only operators with steady laundry or similar workflows have a reason to ask about a pilot. Before signing, ask Dyna for measured MTBI, success rates over a full shift, the named reference site, the price and the service terms. For everyone else, wait for independent footage and published numbers.