Hierarchical Robot Instruction Mapping for Varying Command Granularity

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing robot systems struggle to interpret human-robot instructions of varying granularities efficiently, as they typically assume all commands and tasks reside at a single fixed level of abstraction, leading to inefficient planning and execution in dynamic environments.

Innovation Solution

A system that uses a deep neural network language model to map natural language commands to reward functions at different levels of abstraction within a hierarchical planning framework, enabling the interpretation and execution of commands at multiple levels of specificity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing methods map natural language commands to a formal representation at a fixed level of abstraction, then the robot can complete predefined tasks, but the system becomes unreliable when faced with a changing or stochastic environment

Engineering Contradiction:

Improvereliability in changing environmentVSAvoidadaptability to varying command granularities

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system dynamically adapts the level of abstraction based on the complexity of the natural language command. Instead of using a fixed abstraction level, the hierarchical planning framework allows the robot to select appropriate abstraction levels (high-level for simple commands, low-level for complex commands) in real-time, making the system both reliable in changing environments and adaptable to varying command granularities

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The planning framework is segmented into multiple hierarchical levels of abstraction. Each level handles different types of tasks and commands independently, allowing the system to process complex commands by breaking them down into subtasks at appropriate abstraction levels, thereby improving both reliability and adaptability

Inventive Principle:

Principle #1Segmentation

2Productivity

If existing methods use a fixed level of abstraction for all commands, then the mapping process is simple, but planning and execution times become inefficient due to large state-action spaces

Engineering Contradiction:

Improveplanning and execution efficiencyVSAvoidcomplexity of hierarchical planning framework

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The state-action space is segmented into multiple hierarchical levels, where each level handles a specific scope of tasks. This segmentation reduces the effective state-action space at each level, enabling faster planning and execution while maintaining the ability to handle complex commands through coordinated interaction across levels

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system adds a hierarchical dimension to the planning framework, transforming a two-dimensional state-action space into a multi-layered hierarchical structure. This dimensional change allows the system to navigate complex tasks more efficiently by operating at appropriate abstraction levels rather than searching through the entire state-action space

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS11086938B2Interpreting human-robot instructions

Publication Date: 2021.08.10 BROWN UNIVERSITY

AI summary

A system includes a robot having a module that includes a function for mapping natural language commands of varying complexities to reward functions at different levels of abstraction within a hierarchical planning framework, the function including using a deep neural network language model that learns how to map the natural language commands to reward functions at an appropriate level of the hierarchical planning framework.