1. A Robotic Platform for Exploring Emergent Behavior
Felix Duvallet
Aaron Johnson
Ryan Kellogg
James Kong
Eugene Marinelli
Alexander May
Suresh Nidhiry
Iain Proctor
Gregory Tress
Kevin Woo
Why a colony?
Accomplish tasks that are not feasible with a single robot
Redundancy as a way of achieving robustness
Robots can often fail, especially in dangerous environments
Individual failure will not halt the task
Capabilities of the colony can degrade gracefully
Interesting research questions:
Emergent behaviors
Multi-robot planning and control
Cooperation
Decentralized Simultaneous Localization and Mapping (SLAM)
Abstract
In nature, some of the most successful species are those that form colonies. For instance, while individually vulnerable, myopic, and simple minded, the ant is remarkably effective in groups. We wish to take this same philosophy in designing robots. We seek to explore how complex the behavior of a robot colony can be when the constituent robots are, like the ant, limited in their sensory and processing capabilities yet inexpensive and abundant. To this end, we have developed a platform consisting of a homogeneous group of robots that has enabled us to explore simple behaviors. Furthermore, this platform will enable further research in many robotics topics, including multi-robot localization, large-scale emergent behaviors, communication, and multi-robot cooperation.
Our Colony:
Several homogeneous robotic agents working together to solve a problem
Small size and inexpensive design
Leader-less group
Sensors and wireless communication network
Create complex global behaviors from simple local actions
Robots
Behaviors
Microcontroller
Base Design
Emergent Behavior
Complex global patterns from simple local rules
Each robot has limited sensory and processing capabilities
Multiple simple agents combine to create intricate behaviors
The Firefly+ is the main microprocessor board of each robot
Backlit LCD provides status and debugging information
Easily accessible analog and digital input and output
Multi-colored LED (the “orb”) and other LEDs provide feedback to user and status indications
ATmega 128
Compiles using avr-gcc
Interfaces for sensors and other hardware devices
Serial connection to a computer
User input - two buttons, potentiometer
Light and sound elements
Custom design specifically for the Colony project
Implemented Behaviors
Hunter/Prey
Hunter chases prey
Switch roles when prey is caught
Uses a combination of the BOM, Sharp IR rangefinder, bump sensors, and wireless network
Hunger
Robot simulates hunger, feeding, and emotion
Follow the leader
Robots follow each other
Sensors
Communication
Simulated Behaviors
Bump sensors are attached to the front and rear of the robot
Self-assembling ad-hoc wireless network:
Leader-less: no single robot has more responsibility than the others
Network acts as a token ring, with each robot periodically transmitting
All robots are always listening, but only one is transmitting at a time
Wireless network is closely tied to BOM operation
Bearing data can be linked to wireless data containing robot identification
Each robot is able to gather bearing information for every other robot.
This information can be used in a localization algorithm to create a map of the robot positions
XBee wireless module
Low power, low cost
Sharp IR rangefinder measures distances to objects
Follow the leader
Hunter/Prey
Formation control
The Bearing and Orientation Module (BOM) allows robots to determine relative angular position to other robots
Robots follow one another in a set order
Each robot has a very simple behavior, which combined created an interesting large-scale behavior
One prey (red), many hunters (blue)
Prey moves
Hunters converge on it.
Robots create a shape (in this case a circle)
If any robot is moved, the others will move to compensate and retain formation
BOM Specifications:
Ring of IR emitter-detector pairs
Emitter mode: all emitters powered simultaneously acting as a beacon
Detector mode: each detector polled for analog intensity readings. Most excited detector points in the direction of closest robot in emitter mode
Coplanar among colony robots
Line of sight visibility
Future Behaviors and Applications
Simulation
Acknowledgements
The project was funded in part by Carnegie Mellon’s Undergraduate Research Office and the Ford Motor Corporation. These results represent the views of the authors and not those of Carnegie Mellon University. We would like to thank our advisor Howie Choset, Peggy Martin for her help, as well as Steven Shamlian, Brian Kirby, and Tom Lauwers for their support and contributions to the project.
Simulation can ease behavior development
Enables behavior tests with many more robots
Player/Stage – open-source robot simulator
Additions to simulate custom sensors
In the future, the same code will be used for the simulated behavior and the behavior on the actual robots
More complex formation control and movement
Obstacle detection
Simultaneous Localization and Mapping (SLAM)
Environment mapping and exploration