Dyna Robotics' Taku is a wheeled humanoid built to run end to end chores without a human feeding it. A one hour laundry video is the proof, not the promise.
Dyna Robotics says its Taku robot can run a hotel laundry workflow end to end, without a human feeding it laundry, clearing machines, or resetting the robot when it gets stuck. The company published an uncut one-hour video of the run alongside a technical article on September 29, 2026, the strongest public claim yet that a wheeled humanoid can finish a real chore unsupervised. The story was covered by ABC7 News and the New York Post.
Taku is not a humanoid in the bipedal sense. It has a folding lower body, four steerable wheels for movement, and two seven-degree-of-freedom arms. Dyna is explicit that this is a deliberate trade: a humanoid that can hold a washing machine handle does not need legs if the chore happens on a single floor. The "semi-humanoid" frame is the company's bet that the hard part of a chore robot is not mobility, it is the manipulation and the workflow logic.
Dyna's control stack has three layers. At the bottom, a simulation-trained whole-body controller runs at 100 Hz, the rate at which the robot has to react to its own body and the world. Above that sits DYNA-2, the action policy, and on top, a vision-language orchestrator that handles the workflow: which machine to open, when to fold, when to stop and wait for a human. Dyna calls the top layer a "workflow orchestrator" because laundry is not a single skill, it is a sequence of thirteen decision points, from sorting to loading to folding to interrupting a fold because the dryer has finished.
A single manipulation policy cannot reason about whether the towels are still in the dryer. A workflow orchestrator that reads the room and the schedule can. Dyna also publishes a Unified Robot Representation, a way of describing poses in terms of wrists, elbows, chest, and footprint, that lets the system share data between human demonstration videos and the robot's own trajectories. Joint coordination is left to the controller, which is what makes the human-to-robot data transfer possible without retraining the action policy from scratch.
The video is a marketing artifact; the technical article is the claim. Dyna reports that the run is uncut, one hour, and covers the full workflow including interruptible folding, the robot stops folding a towel when a machine needs to be emptied. The video itself was not independently reviewed here, so the underlying demonstration is being taken on the company's word. That is a normal asymmetry for a vendor-released capability video, and it is why the right read is on the design, not the demo.
There are two numbers in the article that are easy to misread. Dyna reports that stationary folding deployments can reach a production bar in as little as three days, where prior baselines took weeks to months. That is a deployment-time number for a stationary folding system, not a measure of Taku's reliability. Dyna also reports an 87% versus 46% customer acceptance-test result comparing DYNA-2 to DYNA-1, a previous model, on a stationary robot. That is a vendor-reported benchmark on a stationary system, not an end-to-end Taku reliability number. Both figures are useful for context, neither is the headline result.
Figure, 1X, Apptronik, and Sanctuary are all working on humanoid manipulation, and the hard problem for all of them is the same one Dyna is trying to answer: can a humanoid finish a workflow without a human in the loop? Dyna's answer is to remove the legs, double down on arms and workflow logic, and ship a video of one chore. Other teams are betting on legs, on general-purpose foundation models, or on teleoperation fallbacks. The bet is on which design survives a hotel or warehouse at three in the morning when no one is watching.
The honest gap: Dyna has not disclosed pricing, unit availability, or named customer pilots for Taku specifically. The robot is positioned for hotels, restaurants, and warehouses, not homes. The video is one chore, one run, one hotel room. The question is not whether Taku can fold a towel. The question is whether a fleet of Takus can do the next towel, in a different hotel, with a different lighting setup, at 3 a.m. on a Tuesday, without a human watching.