IP Library Granted Patent US 12687855
Granted Patent B2
US 12687855 · App. 18/845,102 · Granted Jul 21, 2026

Bird's-eye data generation device, learning device, bird's-eye data generation program, bird's-eye data generation method, and robot

Inventors: Mai Kurose Nishimura (Tokyo, JP); Shohei Nobuhara (Kyoto, JP); Ko Nishino (Kyoto, JP)
Assignees: OMRON CORPORATION; KYOTO UNIVERSITY
G05D1/24G05D1/646G06T7/20G06T7/70G05D2101/15G05D2111/10G06T2207/20081G06T2207/20084
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12687855
App. No.
18/845,102
Granted
Jul 21, 2026
Kind
B2
Abstract

The present disclosure provides a bird's-eye data generating device with an acquiring section acquiring time-series data; and a generating section generating bird's-eye data.

Claims (35)

1 . A bird's-eye data generating device comprising:

a memory; and

at least one processor coupled to the memory,

the at least one processor being configured to:

acquire time-series data of two-dimensional observation information expressing at least one moving body observed in a dynamic environment from a viewpoint of an observing moving body equipped with an observation device;

generate bird's-eye data expressing an on-ground locus of movement of the observing moving body and on-ground loci of movement of respective moving bodies that are obtained in a case in which the observing moving body is observed from a bird's-eye position, from the time-series data of the two-dimensional observation information and by using a trained model that estimates on-ground motion of the observing moving body and on-ground motions of the respective moving bodies, wherein the trained model includes:

a first encoder whose input is the positions and sizes of the respective moving bodies at a target time, and that outputs vectors,

a second encoder whose input is the on-ground motion of the observing moving body and the on-ground motions of the respective moving bodies that were obtained for one time previously, and that outputs vectors, and

a decoder whose input is the vectors outputted by the first encoder and the vectors outputted by the second encoder, and that outputs on-ground motion of the observing moving body and on-ground motions of the respective moving bodies of the target time; control, by using the generated bird's-eye data, a robot moves to a destination.

2 . The bird's-eye data generating device of claim 1 , wherein

the at least one processor generates bird's-eye data, which expresses the on-ground locus of movement of the observing moving body and loci of movement expressing distributions of on-ground positions of the respective moving bodies that are obtained in a case in which the observing moving body is observed from a bird's-eye position, from the time-series data of the two-dimensional observation information and by using a trained model that estimates the on-ground motion of the observing moving body and distributions of the on-ground motions of the respective moving bodies.

3 . The bird's-eye data generating device of claim 1 , wherein the at least one processor is further configured to track the respective moving bodies from the time-series data of the two-dimensional observation information, and acquire positions and sizes of the respective moving bodies at respective times in the two-dimensional observation information, and

wherein the at least one processor generates the bird's-eye data by using the trained model whose input is positions and sizes of the respective moving bodies at respective times in the two-dimensional observation information and that estimates the on-ground motion of the observing moving body and the on-ground motions of the respective moving bodies.

4 . A non-transitory storage medium storing a program executable by a computer so as to execute bird's-eye data generating processing, the bird's-eye data generating processing including:

acquiring time-series data of two-dimensional observation information expressing at least one moving body observed in a dynamic environment from a viewpoint of an observing moving body equipped with an observation device;

generating bird's-eye data expressing an on-ground locus of movement of the observing moving body and on-ground loci of movement of respective moving bodies that are obtained in a case in which the observing moving body is observed from a bird's-eye position, from the time-series data of the two-dimensional observation information and by using a trained model that estimates on-ground motion of the observing moving body and on-ground motions of the respective moving bodies, wherein the trained model includes:

a first encoder whose input is the positions and sizes of the respective moving bodies at a target time, and that outputs vectors,

a second encoder whose input is the on-ground motion of the observing moving body and the on-ground motions of the respective moving bodies that were obtained for one time previously, and that outputs vectors, and

a decoder whose input is the vectors outputted by the first encoder and the vectors outputted by the second encoder, and that outputs on-ground motion of the observing moving body and on-ground motions of the respective moving bodies of the target time; and

controlling, by using the generated bird's-eye data, a robot moves to a destination.

5 . A bird's-eye data generating method in which a computer executes processing comprising:

acquiring time-series data of two-dimensional observation information expressing at least one moving body observed in a dynamic environment from a viewpoint of an observing moving body equipped with an observation device;

generating bird's-eye data expressing an on-ground locus of movement of the observing moving body and on-ground loci of movement of respective moving bodies that are obtained in a case in which the observing moving body is observed from a bird's-eye position, from the time-series data of the two-dimensional observation information and by using a trained model that estimates on-ground motion of the observing moving body and on-ground motions of the respective moving bodies, wherein the trained model includes:

a first encoder whose input is the positions and sizes of the respective moving bodies at a target time, and that outputs vectors,

a second encoder whose input is the on-ground motion of the observing moving body and the on-ground motions of the respective moving bodies that were obtained for one time previously, and that outputs vectors, and

a decoder whose input is the vectors outputted by the first encoder and the vectors outputted by the second encoder, and that outputs on-ground motion of the observing moving body and on-ground motions of the respective moving bodies of the target time; and

controlling, by using the generated bird's-eye data, a robot moves to a destination.

6 . A robot comprising:

an acquiring section acquiring time-series data of two-dimensional observation information expressing at least one moving body observed in a dynamic environment from a viewpoint of a robot equipped with an observation device;

a generating section generating bird's-eye data expressing an on-ground locus of movement of the robot and on-ground loci of movement of respective moving bodies that are obtained in a case in which the robot is observed from a bird's-eye position, from the time-series data of the two-dimensional observation information and by using a trained model that estimates on-ground motion of the robot and on-ground motions of the respective moving bodies, wherein the trained model includes:

a first encoder whose input is the positions and sizes of the respective moving bodies at a target time, and that outputs vectors,

a second encoder whose input is the on-ground motion of the observing moving body and the on-ground motions of the respective moving bodies that were obtained for one time previously, and that outputs vectors, and

a decoder whose input is the vectors outputted by the first encoder and the vectors outputted by the second encoder, and that outputs on-ground motion of the observing moving body and on-ground motions of the respective moving bodies of the target time;

an autonomous traveling section causing the robot to travel autonomously; and

a control section that, by using the bird's-eye data, controls the autonomous traveling section such that the robot moves to a destination.