Picture a hospital corridor at shift change: nurses pushing carts, a family looking for a room, a doctor walking fast while reading a chart, a patient shuffling with an IV pole. Everyone moves through that space by reading everyone else, stepping aside, making eye contact, waiting for someone to pass.
Now put a robot in the middle of it.
A robot built for the space between people
Diligent Robotics makes Moxi, a hospital robot that handles deliveries such as supplies, medications and lab samples so nurses can spend more time with patients. Moxi does not treat patients. Its job is to move through hospitals safely and get out of the way.
According to The Robot Report, Moxi has been deployed in more than 25 U.S. hospitals. Diligent unveiled Moxi 2.0 in October 2025 and has been rolling it out to hospitals including Endeavor Health Edward Hospital, Providence Saint John's Health Center and Children's Hospital Los Angeles. In January 2026, sidewalk-delivery company Serve Robotics acquired Diligent for $29 million.
The hallway problem
Diligent's CEO has described how the first-generation robot often had to move conservatively in busy hallways. Moxi 2.0 adds compute, sensors and models so it can reason about that complexity in real time.
That phrase, "move conservatively," is worth pausing on. A robot that cannot predict people has only one safe option: slow down and wait. That keeps everyone safe, but it also makes the robot an obstacle. The difference between a robot that works and one that gets in the way is how well it reads people.
Why it matters for Social AI
Navigating a crowded corridor is a social skill. People signal where they are going with their gaze, posture and pace, and they expect others to read those signals. They yield, negotiate and occasionally say "after you."
Robots learn those conventions from data showing how people move and interact in shared spaces, ideally from many viewpoints and with the timing preserved. That kind of data is rare, and in places like hospitals it can only be gathered with careful, documented consent.
Crosstalk sources physical AI data, including collaborative tasks, egocentric video and human–robot interaction, with consent from every person recorded. Explore physical AI datasets or read the paper.
Sources: The Robot Report on Moxi 2.0, CO/AI, Rochester Regional Health.