1.1Fixed camera motion capture
Three synchronized cameras mounted around the work area capture the operator at 60 frames per second. A processor triangulates shoulder, elbow, and wrist positions and converts those positions into robot arm targets.
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.
Three synchronized cameras mounted around the work area capture the operator at 60 frames per second. A processor triangulates shoulder, elbow, and wrist positions and converts those positions into robot arm targets.
A stationary transmitter energizes three orthogonal coils in sequence. A receiver on the operator’s vest measures the resulting fields, and a processor computes torso orientation relative to the transmitter.
Two calibrated cameras track four passive markers arranged in a rigid pattern on the robot’s torso. A processor fits the marker measurements to the known pattern and reports torso orientation in the workcell coordinate frame.
The operator and robot loggers timestamp each sample against a shared clock and write to separate files. At training time, a loader interpolates the robot measurements at each operator timestamp and feeds the aligned values directly to the model.
The logger calibrates both streams into common units and applies the same low-pass filter. It adds calibrated noise to the quieter stream to bring its variance to the measured variance of the noisier stream and rounds both streams to a common step.
Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.
A seated operator wears forearm inertial sensors and elbow angle sensors to guide a humanoid robot that moves padded artifact trays between a cart and an inspection bench. The robot carries corresponding sensors on its forearms and elbows, with angle resolution and stationary noise matched to those of the wearable sensors. The processor maps the operator's motions to robot joint targets, closes the grippers on tray handles in response to a wearable switch, and records timestamped operator and robot measurements alongside the applied commands.
Absolute magnetic encoders replace optical rotary encoders at the wearable elbow hinges and robot elbow output shafts. Each encoder places a diametrically magnetized disk on the rotating member and a two-axis Hall sensor on the stationary member. A calibration sweep supplies offset and gain corrections, and the processor applies a shared angular quantization step to encoder pairs selected for matching stationary noise. The logger associates the corrected angle readings with timestamped inertial measurements and robot commands.
The wearable unit samples its inertial and joint angle channels at 200 samples per second, and the robot samples its corresponding channels at 500 samples per second. Both units timestamp samples against synchronized clocks, and the processor generates actuator setpoints at 200 updates per second from the wearable measurements. A separate logging path low-pass filters both streams at 40 Hz and resamples them onto a common 100 Hz timestamp grid. The logger applies matched quantization steps to corresponding channels and stores the aligned readings, source timestamps, and applied actuator commands.
A humanoid robot loads blanks into a fixture at a guarded milling machine under wearable control. A programmable logic controller exchanges machine-ready, fixture-clamped, and cycle-complete signals with the teleoperation processor. The processor enables loading commands when the controller reports the machine ready for loading, and the controller starts a cycle after it confirms the fixture is clamped, the robot is clear, and the guard is closed. The logger attaches machine-state transitions to the paired wearable and robot sensor records using synchronized timestamps.
A motorized rotary stage carries removable operator-side and robot-side sensors installed in keyed mounts that align corresponding axes. Software controls stationary holds and angular sweeps while recording a reference encoder and both sensor outputs. It estimates axis alignment, bias, scale, quantization step, and noise across the frequency band used by the controller. It stores calibration coefficients and channel acceptance results by sensor serial number, and the teleoperation processor loads the accepted calibration profiles at session start.
The processor continuously records wearable joint measurements, actuator setpoints, and corresponding robot joint measurements on a synchronized time base. For each moving joint, software estimates response delay by correlating commanded and measured joint velocities in overlapping two-second windows. It flags windows with low correlation or a delay above 80 milliseconds and records the estimated delay with the contemporaneous sensor pairs. The training pipeline uses the flags to exclude those windows from batches that require closely tracked motion.
A wearable sleeve carries rigid upper-arm and forearm cuffs connected by an elbow hinge. A magnet and absolute angle sensor measure hinge rotation, and a forearm module measures three-axis acceleration and angular rate.
A brushless motor drives the robot elbow through a reduction gear, and an absolute encoder measures angle at the joint output shaft. A torso-mounted inertial module measures three-axis acceleration and angular rate.
A wearable controller samples joint angles at 100 hertz, and a robot controller maps those angles to joint setpoints within configured travel limits. On each shared clock tick, the recorder stores the wearable and robot measurements in one row with a common timestamp.
The wearable and robot use identical inertial modules configured for the same measurement ranges, sample rate, and filter bandwidth. Bench measurements verify that each inertial pair meets a shared stationary noise tolerance and that each 14-bit joint encoder pair meets shared tolerances for stationary jitter and angular step.
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 motion components 932 include acceleration sensor components (e.g., accelerometer), gravitation sensor components, rotation sensor components (e.g., gyroscope).B · US 12,056,792 B2 · ¶ 127
In one embodiment, the controller may be an infinite-horizon linear quadratic regulator, and an observer may be used to estimate the state of the simplified model based on measurements from the robot. In addition, joint torques of the original model may be sent to joint controllers of the robot being modeled to cause the robot to move.A · US 2014/0249670 A1 · ¶ 14
At step 350, the controller sends commands indicating the joint torques, mapped from the simplified model to the full-dynamics model according to the mapping determined at step 340, to a joint controller capable of configuring the articulated link positions of a humanoid robot by applying the joint torques, thereby causing the robot to move according to the modeled motion.A · US 2014/0249670 A1 · ¶ 52
An empty cell means only that no passage was reproduced there.
Embodiments presented herein provide techniques for simplifying models of robots, and, in particular, for systematically deriving simplified models of humanoid robots.A · US 2014/0249670 A1 · ¶ 2
This approach is difficult to generalize, and it is not always clear how to determine parameters of a simplified model, or if the simplified model captures the essential dynamics of the full model.A · US 2014/0249670 A1 · ¶ 4
More specifically, but not by way of limitation, systems and methods herein describe flow guided transformable bottleneck networks for motion retargeting.B · US 12,056,792 B2 · ¶ 2
A long, recorded video and person-specific training are required for each target person.B · US 12,056,792 B2 · ¶ 3
patent.fyi, US 2026/0355968 A1, claim 1, first published 2026-11-19T18:11:03-05:00, https://patent.fyi/US20260355968/, SHA-256 c8b50cf94c8badf02d6b0679f6768ded2ffce168a2288ab7fd93e15816a99e2d