ORLANDO, Fla. — Home robots are getting better at walking, grasping objects and completing short tasks, but everyday homes remain a difficult test.
A robot in a demo may perform well in a carefully prepared room. The harder challenge is whether it can enter an unfamiliar home, remember where things are and complete a task after the environment changes.
The International Federation of Robotics reported that professional service robot sales reached nearly 200,000 units in 2024, a 9% increase from the previous year. The group has also identified AI-driven autonomy as one of the major robotics trends shaping the industry.
Alper Canberk, founding director of research for robot learning and foundation models at Sunday Robotics, said one of the biggest challenges is teaching robots to understand homes as changing environments, not just spaces to move through.
He said home navigation is often described too narrowly.
“People hear the word and think about movement, path planning, obstacle avoidance, maybe localization,” Canberk said. “Those things matter, but they are not the full difficulty.”
For a household robot, navigation is not just about moving from one point to another. A useful robot may need to clear a table, find the dishwasher, remember where the plates are and continue the task even after objects or rooms leave its camera view.
That requires memory, context and a broader understanding of the home.
Canberk said many robots still perform best in familiar or controlled settings. They can appear capable in places they have seen before, but struggle when moved into a new home with a different layout.
“The difference is whether the model is learning transferable structure or simply exploiting familiarity,” Canberk said.
That distinction matters because real homes are messy. Furniture moves, doors open and close, objects disappear from view and tasks often require more than one step.
Canberk said existing robotic foundation models can be strong over short periods, but many still operate with limited effective memory. That can work for local actions, but not for household jobs that unfold across rooms.
“If a robot is supposed to take a plate from the dining table, find the dishwasher, and place the plate correctly, it needs a durable representation of the home,” Canberk said.
One approach is to train robots with 3D maps of homes and keep those maps available while the robot performs a task. Canberk said his team used 3D room scans across more than 500 homes and collected more than 10,000 hours of navigation data.
The goal was to help the robot rely on the structure of a home rather than memorize familiar layouts.
Canberk said the breakthrough came when the model began using the map as a persistent spatial guide in homes it had not seen before. That allowed the robot to handle tasks that required both navigation and object interaction with more coherence.
Canberk said the difference between a strong demo and a useful home robot comes down to continuity.
“I do not think the winning systems will be the ones that simply move better,” Canberk said. “I think they will be the ones that remember better, interpret better, and integrate action with environmental understanding in a way that persists over time.”
For consumers, that means the future of home robotics may depend less on flashy single-task demonstrations and more on whether robots can understand the ordinary complexity of real homes.
Click here to download our free news, weather and smart TV apps. And click here to stream Channel 9 Eyewitness News live.