Visual-Textual Navigation Agent for Instruction Progress Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In Vision-and-Language Navigation tasks, agents lack explicit representations of navigation goals, making it difficult to determine progress and complete instructions in unknown environments without visual or textual grounding.

Innovation Solution

A self-aware navigation agent with visual-textual co-grounding and a progress monitor module that processes navigation instructions and visual information to identify completed and needed actions, using neural networks for textual and visual grounding, and a sequence-to-sequence architecture with attention to estimate progress towards the goal.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If an agent uses traditional navigation methods without visual-textual grounding, then the system complexity is low, but the agent cannot accurately determine navigation progress or identify completed instructions in unknown environments

Engineering Contradiction:
Improvenavigation progress determination accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent combines visual grounding and textual grounding modules into a unified navigation system. The visual grounding module processes image inputs while the textual grounding module processes language instructions, and both modules work together to determine navigation progress and identify completed instructions, achieving accurate measurement without requiring completely separate systems

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces a progress monitor module as an intermediary that receives inputs from both visual and textual grounding modules. This mediator integrates information from multiple sources to determine navigation progress and identify completed instructions, enabling accurate measurement while managing system complexity through modular design

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If an agent implements comprehensive visual-textual co-grounding with progress monitoring, then the navigation accuracy and instruction completion identification improve, but the computational requirements and processing time increase

Engineering Contradiction:
Improvenavigation reliabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary grounding of visual and textual inputs before navigation decision-making. By pre-processing and grounding the inputs in advance, the system prepares structured representations that can be quickly referenced during navigation, improving reliability while reducing real-time processing delays

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The progress monitor module provides continuous feedback about navigation progress and instruction completion status. This feedback mechanism allows the agent to adjust its behavior based on accumulated information, improving navigation reliability while using efficient iterative processing rather than exhaustive computation at each step

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3814865B1Self-aware visual-textual co-grounded navigation agent
Publication Date: 2022.06.22 SALESFORCE INC
  • EP3814865B1 patent drawingFigure 1
  • EP3814865B1 patent drawingFigure 2
  • EP3814865B1 patent drawingFigure 3

AI summary

An agent for navigating a mobile automated system is disclosed herein. The navigation agent receives a navigation instruction and visual information for one or more observed images. The navigation agent is provided or equipped with self-awareness, which provides or supports the following abilities: identifying which direction to go or proceed by determining the part of the instruction that corresponds to the observed images (visual grounding), and identifying which part of the instruction has been completed or ongoing and which part is potentially needed for the next action selection (textual grounding). In some embodiments, the navigation agent applies regularization to ensures that the grounded instruction can correctly be used to estimate the progress made towards the navigation goal (progress monitoring).