WEBVTT

1
00:00:00.120 --> 00:00:02.590
Welcome to HealthIT.com.

2
00:00:02.590 --> 00:00:04.200
I'm Jivesh Sharma.

3
00:00:04.200 --> 00:00:05.540
A blood draw.

4
00:00:05.540 --> 00:00:07.200
An ultrasound scan.

5
00:00:07.200 --> 00:00:09.630
A delivery down a hospital corridor.

6
00:00:09.630 --> 00:00:16.960
Robotics is taking on very different kinds of work in healthcare, and the last few months have brought developments worth understanding.

7
00:00:16.960 --> 00:00:24.720
In July, the FDA authorized a surgical system with robotic arms integrated into the operating table.

8
00:00:24.840 --> 00:00:29.160
In August came authorization for a supervised blood-draw robot.

9
00:00:29.160 --> 00:00:34.990
In September, researchers reported a robot that acquires thyroid ultrasound scans.

10
00:00:34.990 --> 00:00:41.430
And in October, Google demonstrated an AI model guiding a real robot arm through a physical task.

11
00:00:41.430 --> 00:00:47.120
Let's look at what changed, how the technology works, and where patients and care teams could notice it.

12
00:00:47.686 --> 00:00:50.286
I'm Talia, your AI cohost.

13
00:00:50.286 --> 00:00:52.886
Hospitals have used robots for years.

14
00:00:52.886 --> 00:00:54.528
What makes this wave different?

15
00:00:54.849 --> 00:01:00.269
The change is in how machines interpret a situation and choose their next action.

16
00:01:00.269 --> 00:01:05.349
A carefully programmed movement works well when everything stays in the same place.

17
00:01:05.349 --> 00:01:07.989
A hospital is full of variation.

18
00:01:07.989 --> 00:01:09.609
Someone moves a cart.

19
00:01:09.609 --> 00:01:11.389
A doorway becomes crowded.

20
00:01:11.389 --> 00:01:13.549
A patient shifts position.

21
00:01:13.549 --> 00:01:20.469
Researchers are connecting models that understand images and language with systems that choose physical actions.

22
00:01:20.469 --> 00:01:28.069
Robots can learn from demonstrations, and simulation lets developers practice variations before testing the real machine.

23
00:01:28.069 --> 00:01:35.029
Google's July Gemini Robotics 2 announcement described advances in whole-body control and dexterity.

24
00:01:35.029 --> 00:01:44.509
Its October Argon demonstration made the idea easier to see: a real robotic arm set a fallen mug upright and placed a napkin inside it.

25
00:01:44.509 --> 00:01:49.069
The model used camera observations and a sequence of available actions.

26
00:01:49.069 --> 00:01:53.349
That was a general-robotics demonstration, not a medical procedure.

27
00:01:53.349 --> 00:01:57.709
But watch the pattern: look, act, then look again.

28
00:01:57.709 --> 00:02:00.789
The next move depends on what actually happened.

29
00:02:00.789 --> 00:02:06.141
That feedback is a crucial part of turning AI into a machine that can work around people.

30
00:02:06.451 --> 00:02:11.925
What actually happens between a robot seeing my neck and touching it with an ultrasound probe?

31
00:02:12.305 --> 00:02:16.585
First, it needs to connect what it sees with where its tool is.

32
00:02:16.585 --> 00:02:22.225
A camera can help locate the body; ultrasound can reveal structures beneath the skin.

33
00:02:22.225 --> 00:02:27.105
The robot needs a way to translate between those views and its own coordinates.

34
00:02:27.105 --> 00:02:29.825
That alignment is called registration.

35
00:02:29.825 --> 00:02:34.785
Think of lining up a map with where you are standing before following the directions.

36
00:02:34.785 --> 00:02:36.665
Then comes contact.

37
00:02:36.665 --> 00:02:42.105
The probe needs enough contact to produce an image, with limits on how firmly it presses.

38
00:02:42.265 --> 00:02:44.345
Different devices solve this differently.

39
00:02:44.345 --> 00:02:51.985
An FDA-cleared remote ultrasound system called MELODY, for example, has a specified limit on the force it applies.

40
00:02:51.985 --> 00:02:58.465
The FDA also requires testing of motion accuracy, speed and force for the blood-draw robot.

41
00:02:58.465 --> 00:03:01.305
Finally, the machine needs feedback.

42
00:03:01.305 --> 00:03:04.865
The patient may have shifted, or soft tissue may have moved.

43
00:03:04.865 --> 00:03:10.905
Finding a target once and reaching it accurately throughout a procedure are different engineering problems.

44
00:03:11.065 --> 00:03:15.936
Perception, controlled movement and repeated observation have to work together.

45
00:03:16.056 --> 00:03:20.596
Some of the most practical progress is happening outside the procedure room.

46
00:03:20.596 --> 00:03:27.056
In August, Diligent Robotics announced the rollout of Moxi 2.0 to hospital customers.

47
00:03:27.056 --> 00:03:36.616
The company describes upgraded computing and sensors, learning from fleet experience, and better recovery when the robot encounters an unexpected situation.

48
00:03:36.616 --> 00:03:44.616
In September, NEURA Robotics announced a healthcare business focused initially on moving empty hospital beds and sterile goods.

49
00:03:44.736 --> 00:03:50.296
That is equipment transport, with the robot taking on part of the physical work around care.

50
00:03:50.296 --> 00:03:57.216
These applications matter because a delivery still has to happen when a nurse or technician has something more urgent to do.

51
00:03:57.216 --> 00:04:02.616
A robot that gets a needed item to the right place can potentially give that person time back.

52
00:04:02.616 --> 00:04:08.376
Whether it does depends on the whole job, including loading, handoffs and interruptions.

53
00:04:08.376 --> 00:04:11.256
We often imagine a humanoid at the bedside.

54
00:04:11.326 --> 00:04:15.256
A useful hospital robot may simply be a very dependable courier.

55
00:04:15.629 --> 00:04:18.798
Which of these robots actually touches a patient today?

56
00:04:19.098 --> 00:04:22.258
Aletta is authorized for that specific job.

57
00:04:22.258 --> 00:04:29.378
The FDA authorized it in August to collect blood from adult outpatients under a trained phlebotomist's supervision.

58
00:04:29.378 --> 00:04:36.978
It combines near-infrared imaging with ultrasound, including Doppler information about blood flow, to find a suitable vein.

59
00:04:36.978 --> 00:04:40.538
Without a suitable target, it does not attempt the draw.

60
00:04:40.538 --> 00:04:46.618
If the arm moves too much, the FDA says the needle automatically detaches and the draw stops.

61
00:04:46.678 --> 00:04:51.658
A trained professional starts the session, supervises and handles exceptions.

62
00:04:51.658 --> 00:04:58.567
That is bounded autonomy: the machine performs a defined procedure within a larger service run by people.

63
00:04:58.647 --> 00:05:01.447
Ultrasound shows another important step.

64
00:05:01.447 --> 00:05:12.067
In September, a study in Nature Partner Journals, Digital Medicine evaluated autonomous thyroid scanning in two hundred and sixty-two participants at three institutions.

65
00:05:12.067 --> 00:05:18.007
Here, the robot acquired the images, and separate AI software helped assess them.

66
00:05:18.007 --> 00:05:19.967
That distinction matters.

67
00:05:19.967 --> 00:05:25.567
Software that can interpret a scan still needs someone or something to obtain the right images.

68
00:05:25.647 --> 00:05:33.607
The researchers reported agreement with conventional ultrasound in thyroid measurements, and faster AI-assisted analysis.

69
00:05:33.607 --> 00:05:35.927
Robotic scanning itself took longer.

70
00:05:35.927 --> 00:05:44.407
The combined screening workflow also missed roughly one in five biopsy recommendations, so expert oversight remained important.

71
00:05:44.407 --> 00:05:47.047
Several authors worked for the companies involved.

72
00:05:47.047 --> 00:05:51.647
The technical opportunity is to extend automation to image acquisition.

73
00:05:51.727 --> 00:05:58.767
If that becomes reliable across more patients and settings, imaging could become available in places that struggle to provide it today.

74
00:05:59.054 --> 00:06:01.914
Surgery shows another kind of progress.

75
00:06:01.914 --> 00:06:06.874
In July, the FDA authorized Johnson & Johnson's OTTAVA system.

76
00:06:06.874 --> 00:06:11.994
The company describes four robotic arms built into the operating table itself.

77
00:06:11.994 --> 00:06:18.754
According to the FDA, staff can reposition the patient by adjusting the table without undocking the system.

78
00:06:18.754 --> 00:06:23.354
The company calls the coordinated movement of table and arms twin motion.

79
00:06:23.354 --> 00:06:26.554
The surgeon remains in control of the operation.

80
00:06:26.594 --> 00:06:34.714
The engineering idea is easy to miss: the table, patient and instruments have to work as a coordinated physical system.

81
00:06:34.714 --> 00:06:40.354
Improving that coordination is a different advance from teaching a machine to decide how to operate.

82
00:06:40.354 --> 00:06:46.234
For that more independent future, DARPA launched a surgical robotics competition in September.

83
00:06:46.234 --> 00:06:52.274
One track calls for collaboration with a surgeon; another calls for an independent trauma procedure.

84
00:06:52.274 --> 00:06:55.394
The finals are planned for March 2027.

85
00:06:55.554 --> 00:07:00.663
Those are research targets, including care in settings where surgical help may be scarce.

86
00:07:00.993 --> 00:07:04.083
So when would I actually notice any of this?

87
00:07:04.452 --> 00:07:09.262
Over the next few years, I would watch the ordinary parts of a healthcare visit.

88
00:07:09.262 --> 00:07:11.542
Does the sample reach the lab sooner?

89
00:07:11.542 --> 00:07:15.872
Can a local clinic offer a scan that previously required a long trip?

90
00:07:15.872 --> 00:07:17.942
Can a team add another testing session?

91
00:07:18.502 --> 00:07:22.102
Those are opportunities to test as these systems develop.

92
00:07:22.102 --> 00:07:28.662
They depend on reliable equipment, trained staff and the rest of the service being ready to use the capacity.

93
00:07:28.662 --> 00:07:33.942
As an oncologist, I think about patients who return for blood tests and imaging repeatedly.

94
00:07:34.452 --> 00:07:37.462
Less waiting and fewer journeys would matter to them.

95
00:07:37.462 --> 00:07:41.462
The robot would not need to resemble a doctor for that improvement to be valuable.

96
00:07:42.022 --> 00:07:47.302
Here is the deeper point: clinical capacity can grow one task at a time.

97
00:07:47.302 --> 00:07:50.822
Better AI helps machines interpret a changing world.

98
00:07:50.822 --> 00:07:53.382
Better engineering lets them act in it.

99
00:07:53.382 --> 00:07:59.952
The useful result is a completed delivery, scan or blood draw that helps the next step of care happen.

100
00:07:59.952 --> 00:08:04.022
That is what I will be watching as healthcare AI moves beyond the screen.

101
00:08:04.502 --> 00:08:05.902
Thank you for listening.

102
00:08:05.902 --> 00:08:10.452
If this was useful, follow HealthIT.com for the next conversation.

103
00:08:10.452 --> 00:08:13.782
On YouTube, you can like this episode and subscribe.

104
00:08:13.782 --> 00:08:16.212
And tell us what you'd like us to explore next.

105
00:08:16.212 --> 00:08:20.612
This episode uses an authorized AI-generated version of my voice.

106
00:08:20.612 --> 00:08:24.212
It is general education, not personal medical advice.