Cognitive Colonization

The Robotics Institute

Cognitive ColonizationThe Robotics Institute

Carnegie Mellon University

Dean Boustead, Bernardine Dias, Bruce Digney, Martial Hebert, Bart Nabbe, Tony Stentz, Charlie

Smart, Scott Thayer, Rob Zlot

The Robotics Institute

Cognitive ColonizationNew Ideas

• Free market-based distributed control

• Specialization through functional roles

• Cooperative planning and perception

• Task-level autonomy for robot colonies

Impact• Agile, robust multi-robot systems

• Opportunistic distributed/centralized control

• Sensing scales to mission parameters

• “Fire-and-Forget” mission capability

ScheduleRobust

ColonizationPort to Military

Platforms

ColonizationDynamics

20002000 20012001

20022002

20032003

Static Colonization

The Robotics Institute

Presentation Outline Requirements Software Architecture Multiple Roles Cooperative Stereo Robot Improvements Dynamic Capabilities Experimental Results Status and Future Work

The Robotics Institute

Requirements

is robust to individual robot failure; does not depend on reliable

communications; can perform global tasks given the limited

sensing and computational capabilities of individual robots;

learn to perform better through experience.

Distributed robotics for small-scale mobile robotscalls for a software system that:

The Robotics Institute

Free Market Architecture Robots in a team are organized as an

economy Team mission is best achieved when the

economy maximizes production and minimizes costs

Robots interact with each other to exchange money for tasks to maximize profit

Robots are both self-interested and benevolent, since it is in their self interest to do global good

The Robotics Institute

Distributed Mapping Example

Operator Exec

<-- Revenue paid

Tasks performed -->

The Robotics Institute

Distributed Mapping Roles

Unattached Robot

ReserveRobot

Leader Squad

Mapping Squad

MappingRobot

CommunicationsRelay Robot

The Robotics Institute

Multiple Roles in Simulation

Cost switched from distance based to time based.

Leader and communication roles introduced in addition to sensing role.

Time

Cost

R1

R2

The Robotics Institute

Initial Time-Optimized Plan

The Robotics Institute

Robot Negotiation with Leader

The Robotics Institute

Optimization Via Comms

The Robotics Institute

Robot Improvements I Improved robot dead reckoning capability

by adding gyros. Added “unachievable goal” detection with

re-assignment of tasks to other robots. Rapid deployment through formations

The Robotics Institute

Robot Improvements II Added “dead robot” detection with re-

assignment of tasks to other robots. Added “bump sensing” to cover for

sonar misses: a “bump” puts obstacle in navigation map.

The Robotics Institute

New Dynamic Capabilities

Ported inter-robot negotiation from simulator to real robot test bed to allow for further optimization in response to new tasks, new robots, unexpected results, or lost assets.

Added dynamic goal creation to map an unknown area: four schemes were implemented.

The Robotics Institute

Dynamic Goal Creation Schemes Greedy: robot creates new goals in

unexplored areas near its present location.

Quad-tree: robot creates new goals at center of quad-tree nodes describing unexplored space.

Regular: robot creates new goals in regular pattern over unexplored area.

Random: robot creates random new goals in unexplored area.

The Robotics Institute

Experimental Results

Preliminary results indicate that the random strategy works best because it disperses the robots.

Regardless of the scheme employed, the robots further optimize it by exchanging goals through the negotiation process.

The Robotics Institute

Interior Mapping Video: 4 robots

The Robotics Institute

Interior Mapping Trace

The Robotics Institute

Final Interior Map

45 m

30 m

The Robotics Institute

GUI Map of Interior Environment

The Robotics Institute

Final Exterior Map

The Robotics Institute

GUI of Exterior Environment

The Robotics Institute

Tulip Grove: 5 robots

The Robotics Institute

Cooperative stereo Configuration:

Wide reconfigerable baseline Uncalibrated baseline

Techniques: Robust epipolar estimation Planar homographies

The Robotics Institute

Point correspondences

•Feature seeding•Robust matching (one to many)•Global consistent matching (one to one)•Compute epipolar geometry

The Robotics Institute

Planar correspondences

•Line seeding, 4 cycle generation•Planar matching

•Compute homography•Warp plane•Compute correlation

•Global consistent matching•Reconstruct epipolar geometry

The Robotics Institute

Reconstructed geometry

The Robotics Institute

Current Status Multiple roles tested in simulation. Full control architecture ported to

robots, complete with inter-robot negotiation and robustness to sensing, navigation, and hardware faults.

Distributed mapping demonstrated in unknown environment (interior and exterior) using multiple robots.

The Robotics Institute

Next Steps Switch robots from distance to time cost

regime. Enable robots to sell map information to

each other to improve estimates of navigation costs.

Fuse sonar mapping with stereo mapping to produce human-observable maps of an unknown environment.

The Robotics Institute

Technology Transfer DRES:

NASA:

ARL Robotics CTA: