Hello Robot has launched the fourth generation of its Stretch home-assistance robot, reflecting an approach different from that of companies making broad promises about human-like robots capable of taking on every job. Rather than designing a robot to showcase the greatest possible number of capabilities, the startup is focusing on operating Stretch in homes with real people and building a system that can be controlled safely even when its autonomy is limited.
Hello Robot was founded in 2017 by CEO Aaron Edsinger, a former robotics executive at Google, and CTO Charlie Kemp, a professor at the Georgia Institute of Technology. The company is headquartered in Martinez, California, a city on the northeastern edge of San Francisco Bay, relatively far from the center of robotics companies in Silicon Valley.
A Home Robot That Does Not Try to Fully Imitate Humans
Stretch 4 has a torso with roughly human-like features and a head equipped with several sensors, but it does not fully resemble traditional humanoid robots. Its telescoping arm ends in two grippers, while the robot moves on a heavy base with multidirectional wheels. When its batteries run low, lights around what look like its eyes illuminate, an appearance that company engineer Blaine Matulevich jokingly described as making it look angry.
The new version comes with a limited degree of autonomy, a deliberate decision rather than a shortcoming the company is trying to conceal. Users can give the robot a task and then take over to directly control its movement or arm when necessary. Hello Robot views keeping a human in the control loop as part of the robot’s design, not a flaw. Matulevich said that “control is a feature,” because the user wants control to be part of how the robot embodies itself in the world.
From the Laboratory to the Real Home
Hello Robot’s trajectory suggests that the value of robots is determined not only by the number of tasks they can perform in a demonstration, but also by their ability to operate for long periods in real environments. Advances in artificial intelligence are giving robots greater capabilities, but the sector still lacks sufficient quantities of useful training data. Simulation is improving as well, though investors are increasingly focused on actual deployment.
The article cited a report on the sector published by Bullhound Capital, which said that companies deploying their robots first collect site-specific failure-handling loops and understand workflow limits in each environment—experience that competitors cannot easily buy or generate. From this perspective, robots’ competitive advantage is not limited to intellectual property; it also includes accumulated operating hours in real-world conditions and the responsibility and risks associated with them.
This approach gives Hello Robot an opportunity to develop Stretch in real homes, where spaces, furniture, and daily habits differ from organized laboratory environments. The data generated by these uses can also help improve subsequent robots instead of relying on assumptions that have not been tested by the conditions of everyday life.
Use That Helps People With Mobility Challenges
One of Stretch’s most important applications can be seen in the experience of Keith Platt, an investor in Georgia and a member of Hello Robot’s board of directors. Platt became quadriplegic in 2021 and is no longer able to control anything beyond parts of his shoulders, neck, and head. He began exploring adaptive technologies and then started working with Hello Robot in 2024. The company also includes an occupational therapist to support its work with him and with people facing similar conditions.
Platt controls Stretch through an iPhone app that operates using voice commands. He can ask the robot to move autonomously to a particular place in the house and then take direct control to handle objects and perform tasks. One example he worked on was operating the robot to deliver a protein shake to him in the morning—a task that seems simple but usually requires another person’s assistance.
At the beginning of the trial, Platt needed about two hours to complete the task independently without another person present. But he continued training until he was able to drink the shake and return the cup to the kitchen surface within a few minutes. He also discussed the importance of using Stretch for tasks such as putting on or removing reading glasses and brushing his teeth, explaining that regaining this degree of independence has significant value for him and for the people who care for him.
Platt believes that robotic assistants capable of enabling people with mobility challenges to spend a day safely at home could be a major transformative factor for families, allowing family members to go to work or leave the house without having to employ a professional caregiver all the time. This vision remains tied to the robot’s ability to perform tasks safely, not merely to its having a human appearance.
The Obstacle Is Still Hardware
Stretch’s experience shows that the challenge in robotics is not limited to software and artificial intelligence. Robotic limbs are still heavy and require substantial energy when active balancing is needed. The robotic hand and arm are also much heavier than their human counterparts, while mistakes can damage furniture, appliances, or parts of the home.
Mahi Shafiullah, a postdoctoral researcher working on robotic hands at the University of California, Berkeley, noted that the current state of hardware remains far from the level required to place robots in parents’ homes. He said that industrial robots in his laboratory accidentally pierced a plastic kitchen toy they were supposed to handle carefully. Shafiullah used the third generation of Stretch in his doctoral research at New York University, and models he helped develop using Stretch won the Best Demonstration award at the Computer Vision and Pattern Recognition conference last year, according to the article.
These risks come at a time when some companies have begun marketing more ambitious home robots. The article pointed to 1X, which attracted considerable attention after unveiling its humanoid robot Neo, and said the company had sold all 10,000 units it planned to produce this year, though none had been delivered by the time the article was published.
In contrast, Hello Robot does not promise that Stretch will reach the level of sophistication of humanoid robots attracting Silicon Valley’s attention. Edsinger compares his company’s approach with Waymo, which focused on safety first in developing self-driving cars, while acknowledging that funding was also an important factor. Shafiullah believes Hello Robot is approaching the problem cautiously: It designs the robot to be close to humans first and then looks for capabilities that can be implemented within current safety limits.
Price, Availability, and Customers
Stretch 4 costs about $30,000, a high price for the average consumer but one described as reasonable compared with the price of robots of this type. Edsinger says the price is slightly higher than that of some robots manufactured in China, but those robots generally do not include the sensors or software, which may raise the final cost when these components are added.
Hello Robot expects to manufacture between 200 and 300 units at its headquarters in Martinez, and the first batch has sold out completely. Edsinger also wants to keep Stretch available to researchers and developers with limited budgets. One design criterion is that the robot be shippable inside a cardboard box via UPS or DHL, because switching to wooden crates and installation teams increases the cost and reduces accessibility.
The customer base includes researchers who use Stretch to test more advanced AI models, companies testing its usefulness in environments such as data centers, and people working to develop home assistants for individuals with disabilities. The combination of sensors, motor capabilities, and safe operating procedures gives the robot the potential to play a role in developing what is known as physical AI.
But the path to that runs through data. Shafiullah said that the algorithms may be available, but the data is not, and that data represents about 80% of the influential factor. Therefore, a robot capable of safely collecting this data in the real world could be an important practical step, even if it is less visually exciting than humanoid robots.
Hello Robot intends to continue developing Stretch, with lessons learned from the launch of the fourth-generation robot informing the company’s next robot. This could lead to a lower price and increased capabilities, but the article offers no guarantee that these steps will make human-robot collaboration in homes a widespread reality. What the company’s experience has demonstrated so far is that testing the robot in real homes, while keeping safety and human control at the forefront, provides a different path from making promises greater than the current hardware’s capabilities.