Multi-Agent Pose Estimation with Uncertainty for Dynamic Object Mapping

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current sensor-based systems for robots face limitations in detecting objects outside their field of view and suffer from noisy sensor inputs due to environmental obstructions, and curated maps may not accurately represent dynamic environments with new or changed objects.

Innovation Solution

A method using sensor observations from multiple agents to estimate the pose and uncertainty of objects in an environment by generating a multigraph based on observations, combining measured object poses and uncertainty measures, and applying techniques like spherical linear interpolation and unscented transform to generate composite poses and uncertainty measures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Difficulty of detecting and measuring

If sensor-based detection is used to detect objects, then object detection capability is improved, but detection accuracy deteriorates due to limited field of view and noisy sensor inputs

Engineering Contradiction:
Improveobject detection capabilityVSAvoiddetection accuracy
Core Design Contradiction:
Difficulty of detecting and measuringVSMeasurement precision

Solution Approach 1:

The patent combines data from multiple sensors (cameras, laser scanners, depth sensors) and multiple sources (curated map, sensor inputs) to detect objects. By merging these diverse data sources, the system overcomes the limited field of view and noisy inputs of individual sensors, achieving more accurate and reliable object detection than any single sensor could provide alone.

Inventive Principle:
Principle #5Merging (Combining)

2Ease of operation

If curated map is used for navigation and object detection, then navigation planning is improved, but map accuracy deteriorates due to stale data and lack of dynamic object representation

Engineering Contradiction:
Improvenavigation planningVSAvoidmap accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The system continuously updates the curated map by incorporating real-time sensor inputs from the robot and other agents. This feedback mechanism ensures the map remains current and accurate, reflecting new objects, removed objects, and changed object poses, thereby maintaining high map accuracy while supporting navigation planning.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system performs preliminary actions by proactively detecting and recording changes in the environment (new objects, removed objects, pose changes) before they affect navigation tasks. This ensures the curated map is updated in advance, maintaining accuracy for future navigation and detection operations.

Inventive Principle:
Principle #10Preliminary action

3Reliability

If multiple sensors and data sources are integrated to improve detection accuracy, then object detection reliability is improved, but system complexity increases

Engineering Contradiction:
Improveobject detection reliabilityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces a curated map as an intermediary data structure that organizes and integrates information from multiple sensors and sources. This intermediary layer simplifies the complexity of directly processing all sensor inputs by providing a structured, unified representation of the environment that can be efficiently queried and updated.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3347171B1Using sensor-based observations of agents in an environment to estimate the pose of an object in the environment and to estimate an uncertainty measure for the pose
Publication Date: 2022.11.02 INTRINSIC INNOVATION LLC
  • EP3347171B1 patent drawingFigure 1
  • EP3347171B1 patent drawingFigure 2
  • EP3347171B1 patent drawingFigure 3

AI summary

Methods, apparatus, systems, and computer-readable media are provided for using sensor-based observations from multiple agents (e.g., mobile robots and/or fixed sensors) in an environment to estimate the pose of an object in the environment at a target time and to estimate an uncertainty measure for that pose. Various implementations generate a multigraph based on a group of observations from multiple agents, where the multigraph includes a reference frame node, object nodes, and a plurality edges connecting the nodes. In some implementations, a composite pose and composite uncertainty measure are generated for each of a plurality of simple paths along the edges of the multigraph that connect the reference frame node to a given object node - and a pose and uncertainty measure for an object identifier associated with the given object node is generated based on the composite poses and the composite uncertainty measures.