Multi-Robot Communication

Dr. Daisy Tang

Objectives

� Understand key issues in multi-robot communication

� Understand impact of communication in Balch’s case study with Foraging, Consuming, and Grazing

� Understand how limited communication influences multi-robot coverage behaviors

Multi-Robot Communication

� Objective of communication:� Enable robots to exchange state and

environmental information with a minimum bandwidth requirement

� Issues of particular importance:� Information content

� Explicit vs. Implicit

� Local vs. Global

� Impact of bandwidth restrictions

� Medium: radio, IR, chemical scents, “breadscrumbs”, etc.

� Symbol grounding

Multi-Robot Communication Taxonomy

� By Dudek (1993)

� Communication range:� None, Near, Infinite

� Communication topology:� Broadcast, Addressed, Tree, Graph

� Communication bandwidth:� High (communication is essentially “free”)

� Motion-related (motion and communication costs are about the same)

� Low (communication costs are very high)

� Zero (no communication is available)

Nature of Communication

� One definition of communication:

� “An interaction whereby a signal is generated by an emitter and ‘interpreted’ by a receiver”

� Emission and reception may be separated in space and/or time

� Signaling and interpretation may be innate or learned

� Cooperative communication examples:

� Pheromones laid by ants foraging food

� Time delayed, innate

� Posturing by animals during conflicts/mating etc.

� Separate in space, learned with innate biases

� Writing

� Possibly separated in space & time, mostly learned with innate support

Explicit Communication

� Defined as those actions that have the express goal of transferring information from one to another

� Usually involves:� Intermittent requests

� Status information

� Update of sensory or model information

� Need to determine:� What / When / How / To whom to communicate

� Communications medium has significant impact� Range, bandwidth, rate of failure

Implicit Communication

� Defined as communication “through the world”

� Two primary types:

� Robot senses aspect of world that is a side-effect of another’s actions

� Robot senses another’s actions

Key Considerations in MR Communication

� Is communication needed at all?

� Over what range should communication be permitted?

� What should the information content be?

Is Communication Needed at All?

� Keep in mind:

� Communication is not free and can be unreliable

� In hostile environments, electronic countermeasures may be in effect

� Major roles of communication:

� Synchronization of action: ensuring coordination in task ordering

� Information exchange: sharing different information gained from different perspectives

� Negotiations: who does what?

� Many studies have shown:

� Significantly higher group performance using communication

� However, communication does not always need to be explicit

Over What Range?

� Tacit assumption: wider range is better

� But, not necessarily the case

� Studies have shown: higher communication range can lead to decreased societal performance

Information Content Can Be?

� Research studies have shown:

� Explicit communication improves performance significantly in tasks involving little implicit communication

� Communication is not essential in tasks that include implicit communication

� More complex communication strategies often offer little benefit over basic information

Paper Presentations

� “Communication in reactive multiagent robotic systems”, by Balch and Arkin, Autonomous Robots, 1994.

� “Tenacles: Self-Configuring Robotic Radio Networks in Unknown Environments”, by H. Chiu, et al., 2009.

Introduction

� How should design decisions be made?

� What type, speed, complexity and structure

� Three tasks were devised to discover the impacts of communication

� Performance can be compared across different tasks

� Factors: task, communication type, number of robots, number of attractors, mass of attractors, and percentage of obstacle coverage

Three Tasks: Forage, Consume and Graze

� Mass of attractor determines robot moving speed, consuming speed

� Several robots cooperating can increase the speed

� Size of swath that a robot can graze and the coverage percentage determines grazing time

Task Parameters

� Number of attractors

� Mass of attractors

� Graze coverage

FSAs

In simulation, Graze is implemented by maintaining and marking a high resolution grid corresponding to the environment.

Forms of Inter-Agent Communications

� No communication

� Robots are able to discriminate other robots, obstacles, and attractors

� State communication

� Robots communicate internal states (0: wander, 1: other)

� Behaviors are modified

� Goal communication

� Transmit goal-orientation information

� Behaviors modified accordingly

Explicit vs. Implicit Communication

� The implementation of goal and state communication requires explicit signaling and reception

� Internal states can be observed by other robots

� Robots can also communicate through environment

� Graze

Performance Metric

� Cost

� Reduce cost and minimize # of robots used

� Time

� Maximum # of robots operating without interference

� Energy

� Reliability and survivability

� Greater probability to completion at the expense of time or cost

Environment

� A static flat environment with randomly scattered obstacles

� No a priori map knowledge of obstacles’ location available

� Varied from 5% to 20%

� With 15% as a baseline

Time To Complete

Foraging: Improvement with Communication

Consuming: Improvement with Communication

More On Improvements

Physical Experiments

� Performed on 3 real robots

� Ren, Stimpy and George

� Ren and Stimpy are homogeneous

� Many real world limitations came into play

� Sensors not accurate

� Experimental results still followed simulated results

“Take Home” Message from Balch Communications Study”

� Communication improves performance significantly in tasks with little environmental communication (i.e., with little communication “through the world”, or no stigmergy)

� Communication is not essential in tasks which include implicit communication (i.e., communication “through the world”, or stigmergy)

� More complex communication strategies offer little or no benefit over low-level communication

Summary

� Many types:� Implicit vs. explicit

� Local vs. global

� Iconic vs. symbolic

� Proper approach to communication dependent upon application:� Communication availability

� Range of communication

� Bandwidth limitations

� Language of robots

� Etc.

Tenacles: Self-Configuring Robotic Radio Networks in Unknown Environments

By H. Chiu et al., IROS 2009

Presented by Chris Vigus

Background

� Unknown and dynamic environment

� Unknown locations and distances between agents

� Solution:

� Deploy a group of intelligent robots to explore the environment and position themselves to provide relays

� Robots must self-configure into an effective network, self-heal changes and damages, and adapt to movements of critical entities

Example Problem

Main Contribution

� A distributed coordination algorithm for self-organization and self-healing of robotic networks to establish radio links between critical entities despite the unpredictability and noise of radio signals and unknown locations of entities and relay nodes

� Assumption:

� Critical entities (non-robots) are stationary

Problem Description

� Si and Gi are critical entities to be connected (static)

� Nodes Ni are relay radios

� Goal:

� Nodes move to positions where they relay communication between G and S

� Approach:

� Growing tentacles

� Non-critical tentacles should be removed

Tentacles

� A tentacle consists of a series of stationary robotic radio nodes stretching out from entities

� Entities can communicate when tentacles meet one another

� Useless tentacles (carrying no communication traffic) can be freed and join new tentacle (by detecting radio signal strength)

� Nodes that are part of a tentacle or are relaying network traffic remain relatively stationary

Tentacle Building Algorithm

� Tentacle building

� Tentacle rebuild

� Radio guided exploration

� Local flow optimization

� Node structure:

� nodeID

� A tentacle array (ancestors up to an entity)

� distanceVector

� Boolean flag: tentacle or not

� A probe message is sent periodically to a node’s 1-hop neighbors

� Once a node ceases to receive new messages from its parent, it becomes free

Tentacle Building

� Incrementally connect free nodes to leaf of a tentacle

� A free node joins a tentacle based on:

� Good signal between node and leaf

� No branching. Tentacles may not branch. Multiple tentacles can grow from same entity if children are not within Good signal range

� Weak grandparents

� Avoid previous parents

Tentacle Rebuild

� Evaluating criticality of radio nodes and freeing up stationary nodes

� Non-critical:

� Not connected to other entities; or

� Not essential for carrying traffic between entities

� Closed vs. Open nodes

� Closed: on shortest route from one entity to another

� Determined by distance vectors

� A closed node has its distance vector with at least 2 entries and not dominated

� A rebuild message is initiated from an open leaf and propagates to all ancestors

� A closed node will stop the propagation while an open node

will forward

Radio Guided Exploration

� Goal: move a free node from any region to the boundary of region 4 & 5

Local Flow Optimization

� Local flow optimization mode (LocalOpt) is activated for a node with a low traffic flow

� The node goes forward for a fixed distance with one random direction

� No two neighboring nodes will be in the LocalOpt mode at the same time

Simulation

� Radio signal strength decreases over distance based on an inverse square model

� Strength is fractioned when it penetrates a wall

Simulation Results

� Three initial condition:

� Excellent: radio nodes connect source to gateway with Good link quality

� Fair: source and gateway are initially disconnected and only one radio node next to each entity

� Poor: same as Fair condition, but radio nodes are far away from gateway entity

Self-Healing in Simulation

Experiments on Real Robots

� iCreate platform

� Built-in wall following behavior for exploration

� 7-robot radio network

� 30mx40m environment

Snapshot

Summary

� A bio-inspired tentacle algorithm to stretch tentacle from stationary entities to connect another entity by incrementally connecting radio nodes to leaf of tentacle