NLU Rule Based

Overview

The rule-based Natural Language Understanding (NLU) module is designed to interpret spoken commands by leveraging a fixed set of templates used in RoboCup competitions. By reverse engineering the RoboCup command generator, this module can convert these templates into Regular Expressions (REGEX) patterns. The NLU matches a command to one of its predefined patterns and extracts key information, such as the action verb, objects, and locations mentioned in the instruction.

Features

• Template-Based Command Recognition: Works by matching commands against a fixed set of templates from a pre-built database, ensuring accuracy within the supported set of commands.
• Action Verb and Entity Extraction: Identifies the key elements in a command, such as action verbs (e.g., "find," "pick up") and entities (e.g., objects, locations), using REGEX patterns.
• Structured Command Representation: Transforms recognized commands into a structured format that can be passed to the planner module for further processing.
• Limited to Predefined Commands: The module can only process commands that match existing templates, making it ideal for environments where instructions follow strict patterns but unsuitable for dynamic or novel commands.

Requirements

• ROS version: Noetic
• Dependencies:
  • socrob_speech_msgs
  • socrob_planning_msgs
  • nltk

Installation

0. Install the message modules

Follow the installation instructions in the socrob_speech_msgs and socrob_planning_msgs repositories.

1. Clone the repository

cd ~/<your_workspace>/src
git clone https://github.com/socrob/nlu_rule_based.git

2. Install python dependencies

Ensure that all required dependencies are installed by running:

cd nlu_rule_based
pip install -r requirements.txt

3. Build the workspace

Navigate to your catkin workspace and build the package:

cd ~/<your_workspace>
catkin_make

4. Source the setup file

After building, source the workspace to update the environment:

source ~/<your_workspace>/devel/setup.bash

Usage

Launching the Node

To launch the node, use the following command:

roslaunch nlu_rule_based new_GPSR_nlu_node.launch

This will launch the NLU for the current RoboCup@Home GPSR Command Generator. Other launch files are available:

• GPSR_nlu_node.launch - Launches the NLU for the old GPSR task
• EGPSR_nlu_node.launch - Launches the NLU for the old EGPSR task
• eurobin_coopetition_nlu_node.launch - Launches the NLU for the euROBIN Coopetition.

Launch file arguments

• transcript_topic - Topic for command transcripts. Should publish socrob_speech_msgs/ASRHypothesis message type.
• result_topic - Topic for publishing extracted goals. Uses socrob_planning_msgs/nlu_msg message type.
• use_keyword_nlu - Enables experimental keyword-based NLU. Attempts to infer goals by identifying verbs, objects, and people references if the standard method fails. Recommended for simple use cases only.
• load_from_xml - Loads NLU rules from an XML file if set to true. Should remain false unless using the socrob team's private API.
• asr_error_source - ROS parameter for common ASR errors. Helps correct common speech recognition errors. Should contain mappings of correct spellings to common ASR misspellings. Example:

  pear: [bear]
  crisps: [crisp]
  pringles: [pringle]
  tictac: [tic tac]

  If using the tiago speech recognition module, this parameter will be published to the server when launched.