Over the past few months, it has been quite incredible to see the progress that the robotics field has made. From foundation models to autonomous labs, this rapid rate of development has challenged me to think about what a truly intelligent robot brain might enable.

Specifically, I believe robotics has a real shot at improving the patient experience. From personalized diagnoses to 24/7 autonomous hospitals, I explore what a robotics-powered reshaping of the healthcare system might look like, and how that might inform research directions in the robotics field.

The Diagnostic

Let’s say a patient is experiencing tummy pain. He goes to the hospital, where a nurse performs the standard weight, height, and blood pressure measurements. After being ushered to the doctor’s office, the patient is asked to describe his symptoms. The doctor then proceeds to perform an examination: she uses a stethoscope to hear his heartbeat; presses a hand to his chest to assess his lung capacity; and gently presses his tummy to feel for any alarming symptoms. And accordingly, she recommends that the patient take a blood test.

Though many of these steps don’t necessarily require the specific person performing them, one important role the doctor plays is with the tactile diagnostic — converting physical touch into a tangible diagnosis. This is one of the most important reasons why we go into a hospital in the first place, and why today’s AI models alone are insufficient in medical diagnosis. Touch is an underrated component in unlocking healthcare robotics.

But really, we shouldn’t have to go to a hospital for a simple diagnostic. With the onset of physical intelligence, healthcare diagnosis should be a skill that is included in home-robot pre-training. Our own home robot should be able to feel our tummy and identify any alarming symptoms without having to set foot in a hospital, while additionally synthesizing symptoms and behavior over the past weeks to diagnose.

Towards an Autonomous Hospital

So then, in a world where home robots possess the skills to diagnose and treat common symptoms, what becomes the role of the hospital? The hospital becomes a place where specialized hardware resides; think fully autonomous bimanual blood testers, mobile-robot powered infusion operators, x-ray scanners, and autonomous mammogram conductors, and more.

The most inspiring part of all this is the sheer productivity gains such a system would yield. Right now, the wait time from the identification of a tumor to surgical removal can take up to eight weeks -- an excruciatingly long wait. When hospitals become entirely autonomous treatment centers, surgery can be conducted 24/7, drastically reducing wait times and improving the patient's outcome (both mental and physical) in the process.

From Skill-Constrained → Hardware-Constrained

An interesting artifact of the introduction of autonomous hospitals is that we are no longer constrained by human skill, but rather by the hardware (surgical equipment, transplants, infusion systems) that enables it. The ability to perform surgery is amazingly no longer dependent on the number of surgeons available, but rather the number of robots we have -- every one of these robots can perform surgery since they deploy the same shared intelligence. Importantly, we transition from a healthcare system that is skill-constrained to hardware-constrained.

I believe that this will create a massive demand from the manufacturing industry to produce thousands of robots. The desire for natural resources that facilitate the development of specialized robots will also grow exponentially, putting nations with the ability to acquire these resources at a distinct advantage.

Where does this leave robotics today?

(1) Touch necessitates multimodal models and less invasive forms of data collection

• Current foundation models understandably exclude touch, since a wide variety of tasks can be done using wrist cameras alone. Integration of touch at a large scale has yet to be explored, especially from a healthcare POV.
• But to train large-scale models, we need large-scale data. It’s hard to imagine traditional data collection approaches (two-fingered skill capture mittens or arm puppeteering) for operating specifically with patients. A less invasive approach might include gloves that doctors wear on their hands when diagnosing patients, whose data might then be paired with existing forms of large-scale robot data

(2) Generalist intelligence 🤝 specialized hardware

• We don’t need humanoids to operate blood-drawing, infusion banks, and surgical robots! I believe there’s huge opportunity in creating specialized autonomous medical robots that excel for their specific task — but that they should share a general and powerful intelligence
• Robot foundation models that work across embodiments grow increasingly important here!!

(3) The best home robots will be doctors too