Home automation has spent a decade learning to dim lights and nudge thermostats. At CES 2026, the center of gravity moved to something harder to fake: machines that roll up to appliances, grasp objects, and attempt the untidy sequences of dishwashing and laundry.
That shift matters because domestic work is dominated by the in-between steps that dedicated appliances never solved loading, sorting, unloading, folding, and cleaning up surprises. Zhao Han, assistant professor of human-robot interaction at the University of South Florida, frames the appeal as a morphology problem as much as a software problem: “We have specialized machines like dishwashers and laundry machines, but many chores, in general, require humans,” he says. “These household robots are needed as they take, more or less, human forms with human-like hands and moving capabilities to do those general tasks for us.”
LG’s demonstration of CLOiD leans into the “general tasks” argument by staging full routines rather than single tricks. The company describes CLOiD as a wheeled humanoid with two articulated arms and hands with independently actuated fingers, intended to coordinate tasks across connected appliances. In LG’s own scenarios, the robot retrieves milk from a refrigerator, places a croissant into an oven, starts laundry cycles, and folds and stacks garments after drying. LG also positions its control stack as “Physical AI,” pairing vision-language understanding with action generation trained on household task data.
SwitchBot’s onero H1, by contrast, highlights the gap between a convincing promo clip and the cadence of real manipulation. In booth demos, observers noted the robot taking nearly two minutes to move to a couch, pick up a single clothing item, and place it into a washing machine useful as a proof of concept, but misaligned with the pace of daily life. Even basic interactions, like opening an appliance door, exposed how much reliability hinges on end-effectors, body placement, and battery logistics.
Research benchmarks underscore why “do the laundry” remains a high bar. In EmbodiedBench, multimodal models perform well on high-level instruction tasks but struggle on low-level manipulation; the best reported average score sits at 28.9% on manipulation. A parallel effort, BEHAVIOR-1K, defines 1,000 household activities and shows how chores sprawl into long-horizon plans that require navigation, bimanual handling, and recovery from small errors exactly the failure modes that turn kitchens and laundry rooms into adversarial environments.
Who benefits first is less about novelty and more about friction. Rich Pleeth, CEO and co-founder of Finmile, argues, “Robots are not about laziness,” adding, “Humanity will see huge benefits, particularly those who pay the highest cost in time and physical effort – older adults, disabled people, carers, and time-poor households.” Yet the early systems on show also resemble a new kind of infrastructure: a mobile sensor platform moving through private spaces, seeing and recording by necessity to function.
That is where safety and privacy stop being abstract. ISO’s guidance for personal care robots, ISO 13482:2014, centers on inherently safe design and hazards from human-robot contact issues that become immediate when a tall, heavy device operates around children, pets, glassware, and stairs. Security expert Fergal Glynn calls for minimum data collection, local processing, and straightforward controls to delete recordings and operate offline requirements that will shape trust as much as manipulation accuracy.
For now, CES’ most revealing demos are less about perfect folding than about the engineering bottleneck finally being visible: turning “understand the task” into “execute safely, quickly, and repeatedly” inside a home that was never designed for robots.