Robotic Task Planning for Dependent Natural Language Instructions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current robotic task planning systems struggle to execute complex natural language instructions that include multiple inter-dependent tasks, as they often assume single or serialized tasks, and fail to handle linguistic variations and conditional statements effectively, leading to limited generalization and high annotation requirements.

Innovation Solution

A processor-implemented method using a Conditional Random Field (CRF) model to predict task dependency labels and generate a Conditional Planning Tree (CPT) from complex natural language instructions, allowing for the execution of intended tasks by modifying the original task sequence based on task type labels, arguments, and predicted dependency labels.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If rule-based parsers are used to understand and generate task plans from complex instructions, then the system can handle structured language specifications, but it cannot handle linguistic variations

Engineering Contradiction:
Improvetask plan understanding accuracyVSAvoidlinguistic variation handling
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent replaces rule-based parsers (mechanical/systematic approach) with neural network models that use semantic representations and attention mechanisms to understand natural language instructions, enabling the system to handle linguistic variations while maintaining task plan understanding accuracy

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the parameter representation from rigid rule-based structures to flexible semantic vectors and attention weights, allowing the model to adapt to different linguistic expressions while preserving the core task understanding capabilities

Inventive Principle:
Principle #35Parameter changes

2Productivity

If multiple tasks are embedded in end-to-end training, then the system can plan for multiple tasks, but it is environment specific and does not generalize to new situations

Engineering Contradiction:
Improvemulti-task planning capabilityVSAvoidgeneralization to new situations
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent segments the task planning process into distinct components: semantic representation extraction, task identification, argument extraction, and plan generation. This modular architecture allows each component to be trained independently and combined flexibly for different environments, improving generalization while maintaining multi-task planning capability

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system uses universal semantic representations and task templates that can apply across different environments and task domains. The attention mechanism and semantic role labeling framework provide environment-independent processing that generalizes to new situations while supporting multiple task types

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If significant annotation effort is applied to introduce a new task, then the system can learn the task accurately, but it requires extensive resources and time

Engineering Contradiction:
Improvetask learning accuracyVSAvoidannotation effort and resources
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements self-training and semi-supervised learning mechanisms where the system can automatically generate pseudo-labels and refine its own performance without extensive manual annotation. The pre-trained language model provides initial semantic understanding that reduces the annotation burden for new tasks while maintaining learning accuracy

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system performs preliminary pre-training on large corpora to learn general semantic representations and task patterns before fine-tuning on specific tasks. This preliminary action reduces the annotation effort required for new tasks by transferring learned knowledge, while maintaining task learning accuracy through targeted fine-tuning

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3859587A1Robotic task planning for complex task instructions in natural language
Publication Date: 2021.08.04 TATA CONSULTANCY SERVICES LTD
  • EP3859587A1 patent drawingFigure 1
  • EP3859587A1 patent drawingFigure 2
  • EP3859587A1 patent drawingFigure 3

AI summary

This disclosure provides systems and methods for robotic task planning when a complex task instruction is provided in natural language. Conventionally robotic task planning relies on a single task or multiple independent or serialized tasks in the task instruction. Alternatively, constraints on space of linguistic variations, ambiguity and complexity of the language may be imposed. In the present disclosure, firstly dependencies between multiple tasks are identified. The tasks are then ordered such that a dependent task is always scheduled for planning after a task it is dependent upon. Moreover, repeated tasks are masked. Thus, resolving task dependencies and ordering dependencies, a complex instruction with multiple interdependent tasks in natural language facilitates generation of a viable task execution plan. Systems and methods of the present disclosure finds application in human-robot interactions.