patent.fyi
Applications  /  US 2026/0355944
United States Patent Application Publication

US 2026/0355944 A1

Published
Nov 12, 2026
Filed
May 21, 2024
Earliest date on the face
May 24, 2023
Classification
B25J 9/1612
First published here
Nov 19, 2026

Generated automatically. Not reviewed by a person. Not legal advice or an opinion.

Section 1 · machine-generated

What if?

Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.

1.1Single suction cup wrist

What ifWhat if a single suction cup holds the object while a powered wrist reorients it?
Claim language“a plurality of fingers”

A vacuum pump secures a silicone cup to a smooth face of the object. A pressure switch confirms the seal before a two-axis wrist rotates the cup and object together.

1.1Passive spring fingers

What ifWhat if passive fingers close around the object through spring force and contact geometry?
Claim language“whose state is controllable by a controller”

The wrist pushes two fingers with beveled tips over the rim of a part, spreading them against torsion springs. The springs seat the tips under the rim, and the wrist rotates the retained part. A fixed wedge spreads the fingers after the wrist lowers the part onto a tray.

1.2Magnetic pose sensor

What ifWhat if a magnetic tracker measures the held object's pose relative to the hand?
Claim language“an imaging device”

A three-axis field transmitter on the palm excites its coils in sequence, and a receiver keyed to the object measures the alternating field along three axes. The tracker fits the measured signals to a calibrated field model and reports the object's position and orientation in palm coordinates.

1.2Pose from intensity images

What ifWhat if a monocular camera supplies two-dimensional intensity images of the hand and object?
Claim language“depth images”

The camera records printed markers on the palm and the object in the same grayscale frame. A pose solver uses the known marker dimensions, image corner coordinates, and camera calibration to return the object's pose relative to the palm.

1.3Model-based contact planner

What ifWhat if a geometric planner computes finger commands with a rigid-body model and explicit contact constraints?
Claim language“a machine learning policy computation module”

The controller fits a stored surface model to the measured depth points and computes the rotation to the target pose. An optimizer selects successive fingertip positions and forces subject to friction limits and force balance under gravity. Joint servos execute the resulting sequence while fingertip force sensors maintain the commanded contact loads.

Section 2 · machine-generated

Variations

Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.

Other field of use

In-hand rotation of ceramic insulators

A robotic hand at a ceramic inspection station supports an insulator between three fingers with rubber pads. A fixed depth camera images the fingers and the insulator, and a controller removes points corresponding to the fingers using their measured joint positions. A trained policy uses the remaining depth points to command finger movements that expose successive sections of the insulator to a surface inspection camera while at least two fingers maintain a grip.

Component substitution

Random forest for finger commands

A random forest regressor serves as the controller's policy and predicts an increment for each finger joint. The controller divides the depth image around the grasp into a fixed grid and supplies the median depth and valid pixel count in each cell to the regressor along with measured joint angles. Recorded demonstrations pair these inputs with joint increments for a fixed reorientation task, and the controller executes the predicted increments within joint travel and grip force limits.

Parameter or scale variation

Four-finger hand for small parts

A four-finger hand with 18 mm finger links holds parts measuring 8 to 20 mm across. A depth camera mounted 120 mm from the grasp center supplies images at 60 frames per second, and the policy issues joint position increments at 30 Hz. Joint servos run at 1 kHz and limit each fingertip to 0.8 N of normal force, while the controller permits a finger to reposition only when the other three report contact.

Combination with existing systems

Robot hand feeding a coordinate measuring machine

A robotic arm carries a multi-finger hand between a parts tray and a fixture on a coordinate measuring machine. A wrist-mounted depth camera acquires images of the grasp, and a trained policy commands the fingers to rotate the part while they continue to support it. Once depth measurements locate the part's seating surfaces in the required orientation, the arm seats and releases the part, then signals the measuring machine to start its probe sequence.

Automated or data-driven variant

Automatic slip detection during rotation

Tactile arrays on the fingertips measure normal pressure and tangential load while a depth camera images the held object and the hand. A trained policy uses depth-derived object data to generate finger commands, and a supervisory controller monitors each fingertip for falling normal pressure and rising tangential load. When those readings exceed calibrated slip thresholds, the supervisor pauses the rotation and increases normal force at the supporting fingers within a preset limit. The policy resumes rotation after the tactile readings stabilize and a fresh depth image updates the object pose.

Automatic selection between depth camera views

Two calibrated depth cameras view the grasp from opposite sides, and the controller transforms both point clouds into the hand coordinate frame. Software removes points that coincide with the measured finger geometry and scores each view by the number of remaining points inside the grasp region. At each control update, the policy uses the cloud from the higher-scoring view to issue finger commands, and the controller holds the current grasp when both views fall below a minimum point count.

Specific version of a part

Tendon-driven three-finger hand

Three fingers each have two hinged segments and a tendon routed across both joints. A separate geared motor winds each tendon for flexion, and torsion springs extend the joints as the motor pays out the tendon. Joint encoders report finger angles to the controller.

Active stereo depth camera

Two synchronized cameras view the fingertips and held object while an infrared projector adds a dot pattern to the scene. The processor matches dots across the two images and converts their disparity to depth using the calibrated camera baseline and focal length.

Indirect time-of-flight camera

An infrared emitter illuminates the hand and object with amplitude-modulated light. The sensor measures the phase shift of the return at each pixel and converts it to distance to form a depth image.

Point cloud control policy

The controller extracts the object's depth pixels, projects them into three-dimensional points, and expresses those points in palm coordinates. A neural network trained on simulated manipulation combines the points with joint angles and the target orientation to output incremental joint commands. During rotation, a force loop adjusts finger torque to maintain grip load.

Section 3 · quoted from USPTO records

Other work

Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.

Document AUS 10,059,001 B2
Published
Aug 28, 2018
Filed
Oct 29, 2014
Class
B25J 9/1697
Document BUS 2021/0252700 A1
Published
Aug 19, 2021
Filed
Jan 15, 2021
Class
B25J 9/1607
Claim 1
Document A
Document B
1.1a hand comprising a plurality of fingers whose state is controllable by a controller;
Here, the motion controller 130 moves the two fingers forming the hand 26 again at a predetermined speed in a predetermined direction in a state where the interval between the fingers is farthest opened from each other.
A · US 10,059,001 B2 · ¶ 112
1.2an imaging device for providing depth images of the hand and an object held by the hand; and
In addition, in the present embodiment, for example, a position base method of controlling the robot on the basis of three-dimensional position information on an object calculated using methods, such as a stereogram that uses two sheets of images between which parallax occurs, and makes the image recognize as a solid is adopted as the visual servo.
A · US 10,059,001 B2 · ¶ 98
1.3a controller including a machine learning policy computation module for processing data determined from the depth images of the object held by the hand to provide commands for the hand to reorient the object while maintaining support of the object during the reorientation.
The method includes: constructing an image Jacobian matrix; constructing a depth Jacobian matrix; fusing the image Jacobian matrix and the depth Jacobian matrix, to obtain a hybrid visual servoing Jacobian matrix; and obtaining a robot motion control amount according to the hybrid visual servoing Jacobian matrix by using a control algorithm, to enable a robot to move to a target position, to complete precise positioning.
B · US 2021/0252700 A1 · abstract

An empty cell means only that no passage was reproduced there.

Statements in these documents
Therefore, there is a concern that a hand that grips the object comes into contact with the object unintentionally or the object or the hand itself is damaged.
A · US 10,059,001 B2 · ¶ 8
An advantage of some aspects of the invention is to provide a robot control device, a robot system, and a robot that can reduce unintended contact with an object or a concern that the object or a hand is damaged even if a burden related to calibration is alleviated.
A · US 10,059,001 B2 · ¶ 9
However, problems of information fusion and redundancy of the system need to be solved due to the plurality of sensors.
B · US 2021/0252700 A1 · ¶ 5
The present application discloses a hybrid visual servoing method and related techniques based on fusion of distance space and image feature space, to resolve problems of multi-information fusion and information redundancy, so that a robot can perceive an environment change comprehensively, realize precise positioning of a target object, and complete a servoing task.
B · US 2021/0252700 A1 · ¶ 7
Cite aspatent.fyi, US 2026/0355944 A1, claim 1, first published 2026-11-19T17:33:15-05:00, https://patent.fyi/US20260355944/, SHA-256 10f25a4dc4cff8b9666610be853518f27cbaa4952f6ce03d2984f465f82ab122
First published
Permanent addresshttps://patent.fyi/US20260355944/
Content fingerprint (SHA-256)10f25a4dc4cff8b9666610be853518f27cbaa4952f6ce03d2984f465f82ab122
Methodv2026-11-18.2 · automated, not reviewed by a person
Correctionscorrections@patent.fyi