REACTS 2015

Planning Human-Robot Interaction Tasks Using Graph Models

Introduction

Human-robot interaction is a complex field of robotics in which robots are required to deal with different challenging issues. Among other skills, HRI-capable robots need to generate plans taking humans into account. To achieve this they require sufficiently powerful data structures and rich information not only about their environment, but also about humans and their abilities. A planning-specific requirement when robots are supposed to actively find and classify objects is the capability of creating and retyping symbols –objects– dynamically as a result of the actions of their plans. The work submitted to REACTS 2015 describes how these requirements can be met using a combination of dynamic graph-like world models and a planning system based on graph-rewriting rules. To demonstrate how the approach can be applied, the paper built upon a robot butler use-case, describing how its world model is structured and its most relevant planning rules. Qualitative and quantitative experimental results were also provided.

Domain: full set of planning rules

AGGLPlanner

The AGGLPlanner is part of the AGM project. To install AGGLPlanner:

git clone https://github.com/ljmanso/AGM.git
cd AGM
sh instDep.sh # installs dependencies in Debian-based distributions
make
sudo make install

Figures used in the paper

comparison.png

exampleModel.png

HRI_rule.png

rule0.png

rule15.png

rule22.png