This website is under construction. More updates are on the way.

Sanjay OrugantiResearch Scientist · Rensselaer Polytechnic Institute

← Research

Research project · multi-robot systems

KT-BT

Robots that teach each other new behaviors during a mission

Published in
IEEE Transactions on Robotics, 2023
Presented at
ICRA 2024, Yokohama
My part
First author; part of my Ph.D.
Where
HeRo Lab and DICE Lab, University of Georgia

Overview

In a team of unlike robots, no robot knows how to handle everything. KT-BT (knowledge transfer through behavior trees) lets a robot that meets a situation it cannot handle ask its teammates, receive the missing piece of behavior, and add it to its own behavior tree while the mission is still running.

Framework

Top: three agents exchanging behavior sub-trees through query, response, and update. Bottom: the simulator running a multi-robot search-and-retrieve mission.

Mechanism

Query, response, update

A robot that meets an unknown condition queries its neighbors; a robot that knows the answer responds, and the asker updates its own behavior tree online.

Representation

stringBT grammar

A grammar structure that encodes knowledge as behavior sub-trees, so that behaviors can be shared efficiently between agents.

Evaluation

Search and rescue

Extensive simulations show successful knowledge transfers and improved group performance, and examine how communication range and opportunities affect knowledge spread and functional heterogeneity.

Explore further

Cite

IEEE T-ROAlso presented at ICRA 2024, Yokohama, Japan

KT-BT: A framework for knowledge transfer through behavior trees in multirobot systems

S. Oruganti, R. Parasuraman, and R. Pidaparti

IEEE Transactions on Robotics, 1–17 (2023).