1.1Direct optical gripper tracking
Two calibrated cameras triangulate the centers of three noncollinear markers on the gripper housing. A rigid fit aligns their measured coordinates with their stored coordinates in the gripper frame.
Generated automatically. Not reviewed by a person. Not legal advice or an opinion.
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.
Two calibrated cameras triangulate the centers of three noncollinear markers on the gripper housing. A rigid fit aligns their measured coordinates with their stored coordinates in the gripper frame.
A depth camera mounted above the jaws captures a reference depth map of the open, empty gripper. During operation, a visible jaw marker seeds a connected region of pixels whose depths match that reference map within 2 mm, and the perimeter of the region defines the contour.
The load carries a coating with high near-infrared reflectance, and the jaws have a low-reflectance finish. The processor divides the near-infrared intensity by the visible-band intensity at each registered pixel and labels pixels above a calibrated ratio as load pixels.
Tactile arrays on both jaw faces measure the locations of three raised datum pads on the load, and a jaw encoder supplies the spacing between the arrays. A rigid fit maps the three measured contact points to the corresponding points in a stored load model.
The robot carries the held part to a programmed approach point above a chamfered nest. A wrist force and torque sensor guides lateral and angular corrections during a low-speed descent, and the gripper opens when the robot reaches the programmed seating depth.
Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.
A robot holds a ceramic blank by its unglazed foot while a fixed camera views the blank and part of the gripper against a backlight. The controller computes the gripper pose from the robot joint positions and projects the jaw contact line into the image to form a boundary. It classifies silhouette pixels on the blank side of the line and fits the blank rim to estimate its center and tilt. The controller uses that estimate to set the blank angle and rim height relative to the glaze bath.
A calibrated stereo optical tracker measures robot flange position and orientation in place of joint encoders by triangulating three noncollinear retroreflective markers fixed to the flange. The controller applies a fixed flange-to-gripper transform and projects a plane through the jaw tips as a line in a calibrated inspection image. It uses the line to classify visible corner features on the exposed side of a rectangular load and fits their known geometry to determine the load pose. The robot uses that pose to align the load with a fixture.
A fixed laser line profiler supplies a depth image in place of the area camera while the robot sweeps a held bent tube through the laser plane. The profile rows share a time base with the robot position samples, and the controller maps the intersection of the moving jaw-tip plane and laser plane into each row to form a contour. It selects profile points on the exposed side of the contour and transforms them into gripper coordinates using the pose at each sample time. It fits the straight tube sections and bend to the selected points and commands the robot to align the tube ends with a joining fixture.
Two depth cameras mounted 800 mm apart view opposite ends of a panel up to 1,200 mm wide while a robot grips the panel near its center. The cameras capture synchronized frames at 30 frames per second, and the controller interpolates the robot joint positions to each capture time. In each depth image, the controller projects a gripper-relative cross section as a line and classifies depth samples by side and connectivity to the panel surface. It combines the selected 3D points in robot coordinates, fits a plane and two adjoining panel edges, and uses the fitted pose to align the panel with an assembly frame.
A robot presents a machined part to a coordinate measuring machine equipped with a fixture and a touch probe. A fixed camera calibrated to the robot base images the part and jaw tips, and the controller projects a line through the jaw tips from the robot-derived gripper pose. The line partitions image features by side, and the controller fits known part landmarks from the exposed side to determine the part pose. It uses that pose and the calibrated fixture coordinates to seat the part, then signals the measuring machine to start its probe cycle.
A calibrated camera captures the held part and visible jaw tips, and the controller projects a closed contour around the expected contact patch using the robot-derived gripper pose. Software computes a signed pixel distance from the contour and supplies it as an additional channel to a convolutional network trained on labeled images of held parts. The network labels visible part landmarks, and a perspective pose solver fits their image coordinates to known 3D landmark coordinates. The robot applies the resulting pose to align the part with the fixture.
A six-axis stage transforms the measured load pose into stage coordinates and moves a receiving nest into alignment while the robot dwells at a programmed transfer point. A seat switch on the nest triggers the gripper release.
Absolute rotary encoders report the six joint angles at the camera exposure timestamp. The controller multiplies the calibrated link transforms and the flange-to-gripper transform to calculate the gripper pose in the robot base frame.
The processor transforms a triangle mesh of the jaws into camera coordinates using the determined gripper pose and the measured jaw opening. It projects the mesh through the calibrated camera model and traces the perimeter of the resulting silhouette.
The processor calculates a signed pixel distance from the projected gripper contour, with negative distances inside. It labels pixels more than three pixels inside as gripper, pixels more than three pixels outside as non-gripper, and pixels within the three-pixel band as uncertain.
The processor back-projects depth pixels classified as load into a three-dimensional point cloud. It aligns a stored surface model to the cloud by minimizing point-to-plane distances and returns the resulting rigid transform as the load pose.
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.
The robot controller 50 uses forward kinematics to compute the position of the end effector 44 relative to the second frame of reference 46 associated with the robotic device 40 based on currently measured joint angles.A · US 2016/0214255 A1 · ¶ 17
In at least one embodiment, the pose of the human hand includes a set of articulated segments and joint angles as well as an overall position of the human hand, and the pose of the robotic gripper is determined by duplicating the joint angles of the human hand pose using the robotic hand.B · US 2021/0086364 A1 · ¶ 120
The computer vision steps include computing Fast Point Feature Histogram (FPFH) features for the scene point cloud 35 and the synthesized point cloud model 59 (222), and roughly estimating or otherwise finding the pose of the end effector in the scene employing a Sample Consensus-Initial Alignment (SAC-IA) routine (224).A · US 2016/0214255 A1 · ¶ 38
The partial scene is segmented to a refined scene by a spherical filter with a radius that is equal to the radius of the synthesized point cloud model 59 at the position found by the SAC-IA routine (230). The pose of the end effector in the refined scene is estimated using an Iterative Closest Point (ICP) process starting from the pose found by the most recently executed SAC-IA routine (232).A · US 2016/0214255 A1 · ¶ 38
In at least one embodiment, the uniform sub-sampling used at the input may indicate that points on the fingers are not densely sampled, and therefore a second stage refinement may be needed which resamples points on the hand from the original raw point cloud given the pose and segmentation of the first stage. In at least one embodiment, the second stage may be trained on the same loss functions, but may only use the points sampled on the hand instead to predict accurately the 23 keypoints.B · US 2021/0086364 A1 · ¶ 83
In at least one embodiment, an articulated hand model 310 and the hand pose are processed using DART 312 and kinematic retargeting 314 to produce a corresponding hand pose for a robotic gripper. In at least one embodiment, a control thread applies Reimannian motion policies 318 to the gripper hand pose, and the resulting information is used to control the robot 320.B · US 2021/0086364 A1 · ¶ 68
An empty cell means only that no passage was reproduced there.
This method may be accurate and suitable for a one-time calibration, but can prove tedious for a flexible manufacturing environment where calibration needs to be performed frequently.A · US 2016/0214255 A1 · ¶ 3
Use of a calibration pattern can be problematic in that they may not be suitable for a particular work environment, may disrupt normal operating conditions, and may require position and pattern calibrations, thus increasing the quantity of correspondences and the complexity of the optimization.A · US 2016/0214255 A1 · ¶ 3
At least one embodiment pertains to the use of robots to perform and facilitate tasks under the control of a human operator.B · US 2021/0086364 A1 · ¶ 2
However, such interfaces are generally non-intuitive and difficult to use, requiring significant training and practice to use.B · US 2021/0086364 A1 · ¶ 3
patent.fyi, US 2026/0355969 A1, claim 1, first published 2026-11-19T18:11:03-05:00, https://patent.fyi/US20260355969/, SHA-256 b241ba26da265422f8bcca78175e116847336681e4aacef23878a67d78eed5a9