Granted Patent B2

US 12,364,563 · App. 18/676,049 · Granted Jul 22, 2025

Inventor: Meysam Torabi (Union City, CA)

Assignee: Verb Surgical Inc.

A61B34/75A61B34/25A61B34/35A61B34/74G06N3/04G06N3/044G06N3/08G16H40/67A61B2034/301A61B2034/302

Abstract

Claims (74)

1. A computer-implemented method for detecting an intention to disengage from a surgical tool in a robotic surgical system, the method comprising:

receiving a plurality of time-series which include i) a first time-series of positions of a first user input device, UID, controlling a movement of a first surgical tool and ii) a second time-series of positions of a second UID controlling a movement of a second surgical tool;

applying the plurality of time-series to a neural network classifier, wherein the neural network classifier is trained to detect disengagement and teleoperation motions of the first UID and the second UID and generate either a disengagement decision or a teleoperation decision; and

in response to receiving the disengagement decision from the neural network classifier, disengaging the first UID and the second UID from controlling the first and second surgical tools.

2. The computer-implemented method of claim 1 , wherein each of the first and second time-series of positions of the first UID and the second UID is a time-series of three-dimensional, 3D, positions of a respective UID.

3. The computer-implemented method of claim 2 , further comprising:

computing a derivative of the first time-series to generate a first derivative time-series of the first UID representative of a velocity of the first UID;

computing the derivative of the second time-series to generate a second derivative time-series of the second UID representative of a velocity of the second UID; and

applying the first derivative time-series and the second derivative time-series as input to the neural network classifier.

4. The computer-implemented method of claim 3 , further comprising:

computing the derivative of the first derivative time-series to generate a third derivative time-series of the first UID representative of an acceleration of the first UID; and

computing the derivative of the second derivative time-series to generate a fourth derivative time-series of the second UID representative of an acceleration of the second UID, and wherein the neural network classifier is configured to receive as input the third derivative time-series and the fourth derivative time-series.

5. The computer-implemented method of claim 4 , further comprising:

computing the derivative of the third derivative time-series to generate a fifth derivative time-series of the first UID representative of a jerk of the first UID; and

computing the derivative of the fourth derivative time-series to generate a sixth derivative time-series of the second UID representative of a jerk of the second UID, and wherein the neural network classifier is configured to receive as input the fifth derivative time-series and the sixth derivative time-series.

6. The computer-implemented method of claim 5 , further comprising:

computing the derivative of the fifth derivative time-series to generate a seventh derivative time-series of the first UID representative of a snap of the first UID; and

computing the derivative of the sixth derivative time-series to generate an eighth derivative time-series of the second UID representative of a snap of the second UID, wherein the neural network classifier is configured to receive as input the seventh derivative time-series and the eighth derivative time-series.

7. The computer-implemented method of claim 3 , further comprising:

computing a ninth time-series by summing the first derivative time-series and the second derivative time-series representative of combined velocities of the first UID and the second UID;

computing a tenth time-series by performing a subtraction between the first derivative time-series and the second derivative time-series representative of differences in velocities between the first UID and the second UID;

computing the derivative of the ninth time-series to generate an eleventh derivative time-series representative of a rate of change of the combined velocities; and

computing the derivative of the tenth time-series to generate a twelfth derivative time-series representative of a rate of change of the differences in velocities, wherein the neural network classifier is configured to receive as input the eleventh derivative time-series and the twelfth derivative time-series.

8. The computer-implemented method of claim 3 , further comprising:

generating a distance time-series by computing a set of distance values between each 3D position in the first derivative time-series and a corresponding 3D position in the second derivative time-series representative of distances between the first UID and the second UID;

computing the derivative of the distance time-series to generate a fifteenth derivative time-series representative of a rate of change of the distances between the first UID and the second UID;

computing the derivative of the fifteenth derivative time-series to generate a sixteenth derivative time-series representative of a rate of change of the fifteenth derivative time-series; and

computing the derivative of the sixteenth derivative time-series to generate a seventeenth derivative time-series representative of a rate of change of the sixteenth derivative time-series.

9. The computer-implemented method of claim 3 , wherein applying the first derivative time-series and the second derivative time-series as input to the neural network classifier includes:

applying a sliding window of a predetermined window size to each of the first derivative time-series to generate a composite data sample comprising a set of data points from each of the first derivative time-series within the sliding window; and

applying the neural network classifier on the composite data sample to generate either the disengagement decision or the teleoperation decision based on the composite data sample.

10. The computer-implemented method of claim 1 , wherein prior to applying the plurality of time-series to the neural network classifier, the method further comprises training the neural network classifier by:

receiving a first plurality of time-series of positions of a left UID and a second plurality of time-series of positions of a right UID;

for each time-series in the first plurality of time-series and a corresponding time-series in the second plurality of time-series:

annotating the time-series to identify each teleoperation event in the time-series and each UID docking event in the time-series; and

processing the annotated time-series to generate a plurality of derivative time-series of the first plurality of time-series and the second plurality of time-series; and

using the plurality of derivate time-series of the first plurality of time-series and the second plurality of time-series as training data to train the neural network classifier.

11. The computer-implemented method of claim 10 , wherein annotating the time-series to identify each teleoperation event in the time-series includes:

identifying a first timestamp corresponding to the beginning of a teleoperation event and a second timestamp corresponding to the end of the teleoperation event; and

labeling each data point between the first timestamp and the second timestamp with a first value indicative of a teleoperation mode.

12. The computer-implemented method of claim 11 , wherein annotating the time-series to identify each UID docking event in the time-series includes:

identifying a third timestamp corresponding to the beginning of a UID docked mode; and

labeling each data point between the second timestamp and the third timestamp with a second value indicative of a disengagement mode, wherein the disengagement mode terminates when the UID docked mode begins.

13. The computer-implemented method of claim 1 , wherein the disengagement motions include docking motions of returning the first and second UIDs to respective docking stations.

14. The computer-implemented method of claim 1 , wherein the neural network classifier includes a plurality of LSTM layers followed by a dropout layer with a dropout rate between 0.3-0.4, e.g., a dropout rate=0.3.

15. An apparatus for detecting a disengagement intention from a surgical tool in a robotic surgical system, the apparatus comprising:

one or more processors; and

a memory coupled to the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the apparatus to:

receive a first time-series of positions of a first user input device (UID) controlling a movement of a first surgical tool and a second time-series of positions of a second user input device (UID) controlling a movement of a second surgical tool;

process the first time-series and the second time-series to generate a plurality of derivative time-series of the first time-series and the second time-series;

apply a neural network classifier on the plurality of derivative time-series, wherein the neural network classifier is trained to detect disengagement and teleoperation motions of the first and second UIDs and generate either a disengagement decision or a teleoperation decision; and

in response to receiving the disengagement decision from the neural network classifier, disengaging the first and second UIDs from controlling the first and second surgical tools.

16. The apparatus of claim 15 , wherein the memory further stores instructions that, when executed by the one or more processors, cause the apparatus to generate the plurality of derivative time-series by:

computing a derivative of the first time-series to generate a first derivative time-series of the first UID representative of a velocity of the first UID; and

computing the derivative of the second time-series to generate a second derivative time-series of the second UID representative of a velocity of the second UID.

17. The apparatus of claim 16 , wherein the memory further stores instructions that, when executed by the one or more processors, cause the apparatus to apply the neural network classifier to the plurality of derivative time-series by:

applying a sliding window of a predetermined window size to each of the first derivative time-series and the second derivative time-series to generate a composite data sample comprising a set of data points within the sliding window; and

applying the neural network classifier on the composite data sample to generate either a disengagement decision or a teleoperation decision based on the composite data sample.

18. The apparatus of claim 15 , wherein the memory further stores instructions that, when executed by the one or more processors, cause the apparatus to train the neural network classifier by:

receiving a first plurality of time-series of positions of a left UID and a second plurality of time-series of positions of a right UID;

for each time-series in the first plurality of time-series and a corresponding time-series in the second plurality of time-series:

annotating the time-series to identify each teleoperation event in the time-series and each UID disengagement event in the time-series; and

processing the annotated time-series to generate a plurality of derivative time-series of the first time-series and the second time-series; and

using the annotated and processed derivatives of first plurality of time-series and the second plurality of time-series as training data to train the neural network classifier.

19. A robotic surgical system, comprising:

a first and a second surgical tools;

a first and a second user interface device (UID) for controlling movements of the first and the second surgical tools, respectively; and

a computer coupled to the first and the second surgical tools and the first and the second UIDs and configured to perform real-time UID disengagement detections by:

receiving a first time-series of positions of the first UID and a second time-series of positions of the second UID;

processing the first time-series and the second time-series to generate a plurality of derivative time-series of the first time-series and the second time-series;

applying a neural network classifier on the plurality of derivative time-series, wherein the neural network classifier is trained to detect disengagement motions and teleoperation motions of the first and second UIDs and generate either a disengagement decision or a teleoperation decision; and

in response to receiving the disengagement decision from the neural network classifier, disengaging the first and second UIDs from controlling the first and second surgical tools.

20. The robotic surgical system of claim 19 , wherein the plurality of derivative time-series include at least the velocities and accelerations of the first time-series and the second time-series, and wherein applying the neural network classifier on the plurality of derivative time-series allows for significantly reducing the data size required to make each disengagement or teleoperation decision, thereby expediting the speed of performing the real-time UID disengagement detections and preventing unintended movements of the surgical tools.

21. The robotic surgical system of claim 19 , wherein the disengagement motions include docking motions of returning the first and second UIDs to respective docking stations.

Assignments (1)

Continuity (2)

References Cited (33)