patent.fyi
Applications  /  US 2026/0355947
United States Patent Application Publication

US 2026/0355947 A1

Published
Nov 12, 2026
Filed
May 6, 2025
Earliest date on the face
May 6, 2025
Classification
B25J 9/163
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.

1.1Continuous scene variables

What ifWhat if the planner represents the scene with continuous measurements and evaluates candidate motions by their numerical costs?
Claim language“a predicate library”

The scene estimator stores object poses, gripper aperture, and signed distances to nearby surfaces as floating-point values. The planner scores candidate motions using final pose error and clearance penalties computed from those values.

1.2Simulated action trials

What ifWhat if a physics simulator supplies the action outcomes used to build the action library?
Claim language“a human demonstration”

A trial generator varies gripper approach angle, closing force, and workpiece pose in a rigid-body simulator. A rule extractor evaluates the predicate library before and after each trial and records recurring preconditions and effects for successful commands.

1.2Direct replay of recorded motion

What ifWhat if the controller executes a single recorded motion with position and force feedback?
Claim language“an action library”

The recorder logs time-stamped tool poses and gripper commands during a demonstration. At run time, the controller transforms the recorded poses into the current fixture frame and pauses playback when a wrist force sensor exceeds a set threshold.

1.3Runtime library lookup

What ifWhat if the planning domain uses live references to both libraries and the planner reads their entries during search?
Claim language“compiling the predicate library and the action library”

A task configuration stores the addresses of the predicate and action libraries in separate table stores. During search, the planner retrieves applicable action entries by object type and calls the referenced predicate evaluators against the current state.

1.3Fixed logical vocabulary

What ifWhat if the planning domain requires both predefined predicates and predefined actions?
Claim language“without requiring predefined logical predicates or predefined logical actions”

A fixed schema supplies predicates for gripper occupancy and fixture occupancy, along with grasp and release actions that specify their preconditions and effects. A configuration step binds the schema's parameters to detected workpieces, fixture pockets, and robot grippers.

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

Planning for reusable tote washing

A tote washing station takes a cleaning procedure and video of an operator loading, washing, draining, and stacking reusable totes. A language model proposes predicates for tote orientation, rack occupancy, and wash completion from the procedure and observed station states. The generator evaluates those predicates using camera detections and washer cycle logs, then derives action preconditions and effects from the operator's recorded runs. A compiler joins the predicates and actions in a planning domain for a tote handling manipulator.

Component substitution

Inductive logic programming for predicate generation

An inductive logic programming module replaces the language model in the predicate generator. A task description supplies object types and numeric acceptance limits, and successful and unsuccessful demonstration checkpoints supply labeled examples over measured positions, contact events, and switch readings. The module searches bounded conjunctions of comparisons and relations to produce predicate definitions that separate the labeled examples. The action generator evaluates those definitions at demonstration checkpoints, infers preconditions and effects, and compiles both libraries into the planning domain.

Parameter or scale variation

Generation across eight assembly benches

Eight assembly benches use wrist cameras and fixture sensors to record 20 human demonstrations per bench at 10 samples per second. The predicate generator uses the shared task description and pooled observations to define state properties with bench, fixture, and part identifiers as arguments. The action generator groups demonstration transitions by the predicates that change and retains shared preconditions that hold at every observed start within each group. The compiler emits one planning domain with bench and fixture arguments so that a new task can bind actions to a particular bench.

Combination with existing systems

Integration with a warehouse control system

A planning domain generator connects to a warehouse control system through a work-order queue and an equipment status feed. It uses work-order descriptions and synchronized recordings of human pick-and-place demonstrations to generate predicates for bin occupancy, item identity, and carrier position, along with actions that change those states. A compiler attaches an equipment command mapping to each generated action and supplies the domain to the warehouse task planner. The warehouse controller issues each command and checks its completion event against the expected state change before dispatching the next action.

Automated or data-driven variant

Sensor based segmentation of demonstrations

A handheld demonstration gripper records jaw position and grip force while an overhead camera tracks the tool and workpieces. The predicate generator uses the task description and synchronized sensor records to define properties such as object held, object supported, and fixture occupied. A multivariate change-point detector identifies candidate action boundaries from force and motion changes, and the action generator compares stable predicate values on either side of each boundary to derive action effects. The compiler checks each effect against the predicate library and writes both libraries into the same planning domain.

Predicate refinement from failed demonstrations

A domain generator reads a task description and demonstration recordings labeled by whether the operator completes each operation. It replays failed attempts through the generated action definitions and groups cases where a successful attempt and a failed attempt share the same recorded logical preconditions. A decision tree searches recorded measurements such as insertion depth and angular alignment for thresholds that separate the two groups. The generator adds the resulting threshold predicates to the library, updates the relevant action preconditions, and recompiles the planning domain.

Specific version of a part

Geometry-based predicate generation

A language model extracts object roles and spatial relationships from an instruction to seat a pin in a fixture bore. The generator registers a seating predicate whose evaluator checks lateral offset below 0.5 millimeters and insertion depth within 0.5 millimeters of the target.

Contact-segmented action extraction

A recorder captures tool pose, gripper aperture, and wrist force during a human demonstration. A segmenter detects gripper transitions and sustained contact, then an extractor compares predicate values at segment boundaries to record action preconditions and effects.

Typed planning domain compilation

A compiler assigns object types to parameters in the generated predicate and action entries, then resolves each precondition and effect against a predicate signature. It writes the declarations and transition rules into one domain file for a symbolic planner.

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 2023/0084774 A1
Published
Mar 16, 2023
Filed
Sep 14, 2022
Class
B25J 9/1664
Document BUS 2025/0005318 A1
Published
Jan 2, 2025
Filed
Jul 1, 2024
Class
G06N 3/006
Claim 1
Document A
Document B
1.1generating a predicate library comprising logical predicates defining properties or relationships of objects relevant to a task description;
A set of predicates that describe a domain of the target system is determined from the received logs. The set of predicates represents an initial state and sequence state changes, while each predicate applies to one or more objects in the domain.
B · US 2025/0005318 A1 · ¶ 7
1.2generating an action library comprising logical actions defining transitions between logical states based on a human demonstration interpretable using the predicate library; and
In some examples, an operator (e.g., a remote human operator) may be tasked with manually controlling the perception system based on data received from the robotic device. The trajectory of the perception system as controlled by the remote operator and various sensor data from the robotic device may be stored and used to train a machine learning model that can output perception system trajectories.
A · US 2023/0084774 A1 · ¶ 33
During domain synthesis, definitions, pre-processed predicates, user-defined input, existing models, and other information, such as raw logs, are input to synthesis a model that describes the actions that lead from one system state to another. In the game example, domain synthesis can take log data from a human player and develop a PDDL model that solves any Towers of Hanoi puzzle.
B · US 2025/0005318 A1 · ¶ 52
1.3generating the planning domain based on compiling the predicate library and the action library, the planning domain being generated without requiring predefined logical predicates or predefined logical actions.
Incremental auto synthesis (step 13) is performed to generate a domain model, such as in PDDL format, that describes the log data. The auto synthesis can use an empty model or an existing model.
B · US 2025/0005318 A1 · ¶ 31

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

Statements in these documents
As robotic systems become increasingly prevalent in numerous aspects of modern life, it is desirable for robotic systems to be efficient.
A · US 2023/0084774 A1 · ¶ 2
Robotic devices may be used for applications involving material handling, transportation, welding, assembly, and dispensing, among others.
A · US 2023/0084774 A1 · ¶ 2
Therefore, a need remains for a data-driven approach that takes operational logs from a system and creates a planning domain model in a modeling language.
B · US 2025/0005318 A1 · ¶ 3
In conjunction with the previous logs and the previous model, a new model can be synthesized more efficiently and will be more accurate.
B · US 2025/0005318 A1 · ¶ 6
Cite aspatent.fyi, US 2026/0355947 A1, claim 1, first published 2026-11-19T17:33:15-05:00, https://patent.fyi/US20260355947/, SHA-256 c0fc21e1bf6489fab464e2ea9ca4565632a4e27e6570a0ae7cf672634835e589
First published
Permanent addresshttps://patent.fyi/US20260355947/
Content fingerprint (SHA-256)c0fc21e1bf6489fab464e2ea9ca4565632a4e27e6570a0ae7cf672634835e589
Methodv2026-11-18.2 · automated, not reviewed by a person
Correctionscorrections@patent.fyi