Eye surgery robots: the hard part isn't precision, it's feel

As someone who knows nothing about ophthalmology, I tried ForSight Robotics' JASPER at a booth at a medical equipment expo.

First, the conditions. You can't take this thing home, and there's no way it's letting me operate on a real human eye. The engineer turned on simulation mode, and I operated on an artificial eyeball model for about ten minutes, watching two rounds of demos before and after. So what follows is about the feel and the workflow impressions, not clinical conclusions—don't use it as reference.

My judgment is that this thing isn't sold to individual doctors; it's for institutions willing to rebuild an entire surgical workflow. If all you want to know is whether the robot stitches more accurately than a human, you're asking the wrong question.

First, let me translate a few terms. Cataract surgery is replacing the clouded lens in the eye with an artificial one, with incisions down to the millimeter level, all done under a microscope. Master-slave teleoperation means the doctor doesn't hold the instruments directly—they sit at a console pushing handles, and the robot executes on the patient's eye, with algorithms and servos in between. JASPER is the latter, made by an Israeli company, formerly called Oryom.

I sat down, and the first thing I felt was the field of view. Under a traditional microscope the doctor's eyes are pressed to the eyepieces, and the view shakes whenever the head moves. This one uses a screen-based 3D display, and the image is stable. The engineer overlaid a real-time guidance layer marking where to cut. I think this is genuinely useful, especially for doctors who don't do many of these. The official line is that it shortens the learning curve, with real-time intraoperative guidance aiding decisions.

Then I got hands-on. Push the handle, and the end instrument moves. I didn't detect obvious lag on my end, but there was one thing I wasn't prepared for: my movements were amplified, and also filtered. My hand shakes—it doesn't; I want to go fast—it lags half a beat. For the first five minutes I was fighting myself.

The engineer had me do a circular capsulorhexis on the model—tearing open a thin membrane along a circle. I did it six times, succeeded once, and even that was crooked.

This isn't a product problem. It actually shows that "shortens the learning curve" has a precondition: it shortens the curve for ophthalmologists, not for someone like me. Your existing feel has to be torn down and rebuilt, and nobody saves you that cost.

That later number is even more telling. JASPER has completed 26 robotic cataract surgeries with zero conversions back to manual.

26 cases isn't many. But zero manual conversions is a hard metric. Anyone who does robotics knows that running through once and running through consecutively are two different things—one mid-procedure takeover means the workflow still has a break point. 26 cases with no takeover means at least at the workflow level they've thought it through.

There are gripes too. I basically couldn't feel any force feedback—the whole time I relied on visual judgment of whether tissue was being pulled. That's not new in master-slave surgical robots, but for an organ like the eye, I think the sense of touch will have to be added sooner or later. Also, the console's footprint and the OR renovation aren't small costs—they didn't say how much, and I won't guess.

So is it suitable or not. Ophthalmology centers with steady surgical volume, willing to send doctors for long-cycle training, and able to rearrange their ORs should get in touch now—the earlier you're in, the more you accumulate. Individual doctors, small institutions, or anyone who wants to buy one and have it work automatically—not suitable; the value of this thing is in the workflow, not the standalone machine.

As for these platforms all saying "let doctors do better"—the medical accounting doesn't work that way. The extra equipment and training costs ultimately land on the department, the hospital, or the patient, and whether payers accept that premium. 26 cases prove the technology runs; they don't prove the business runs. That's what I want to see more.