implementing consensus based tasks with autonomous agents...

42
The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 78249 9/2/2016 1 Implementing Consensus based tasks with autonomous agents cooperating in dynamic missions using Subsumption Architecture Prasanna Kolar PhD Candidate, Autonomous Control Engineering Labs University of Texas at San Antonio Copyright Prasanna Kolar, UTSA

Upload: others

Post on 25-Mar-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 1

Implementing Consensus based tasks with autonomous agents cooperating in dynamic

missions using Subsumption Architecture

Prasanna Kolar PhD Candidate,

Autonomous Control Engineering LabsUniversity of Texas at San Antonio

Copyright Prasanna Kolar, UTSA

Page 2: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 2

Outline

Challenge

Contributions

Introduction

Experimental Details

Results and Discussions

Conclusion

Future Work

Acknowledgements

References

Copyright Prasanna Kolar, UTSA

Page 3: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 3

Challenge

• Develop a Robust Task Allocation & Task Execution unmanned system Architecture by integrating a decentralized task allocation system with a decentralized behavior task execution system

• We Integrated Consensus Based Bundled Algorithm(CBBA) with Subsumption Architecture(SA)

Why?

• Modern day systems are network oriented

• Multi-application systems

• Need for proper coordination and collaborationto successfully execute missions

• Systems work in areas that are unknown or partially known environments

Copyright Prasanna Kolar, UTSA

Page 4: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 4

Contributions

• Development of an UAV System architecture by integrating CBBA with Subsumption Architecture

• Implemented the above contributions in simulation and on hardware

• Research and Development of Bound CBBA (BCBBA) which enables faster bidding up to 20%

Copyright Prasanna Kolar, UTSA

Page 5: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 5

Introduction .. Centralized systems

Multiple child system transactions are handled by one leader system

One main system handles most of the calculatio

Others just follow its ‘orders’

Advantages:

powerful hardware for computation only in main system

Sub systems can have inexpensive hardware

Disadvantages:

Single system decision making

Single Point of Failure

Heavy network traffic between main and sub systems

Centralized System

Copyright Prasanna Kolar, UTSA

Page 6: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 6

Introduction .. De-Centralized systems

• All systems are decision makers using consensus

• Local information used to achieve global goals

Advantages

• Faster and more up-to-date information

• Less network traffic

• No single Point of Failure Hence system continues even if some drones fail

Disadvantages

• Complex logic for consensus strategies

• The drones may need more expensive hardware

Well connected Decentralized system

Copyright Prasanna Kolar, UTSA

Page 7: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 7

Consensus – Are we in agreement? !!

• Agent agreement

• Information sharing and auction

• Robust to network topologies

Auction – My bid is larger than yours!

• Calculate bids using predefined scoring logic• Robust to inconsistencies in situation awareness

Combine both?

Task Selection – use AuctionConflict Resolution – use Consensus

Task Allocation

Copyright Prasanna Kolar, UTSA

Page 8: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 8

CBBA Logic

Task Selection

• Bundle the tasks Calculate Bid

Choose Tasks that maximize the total score

Conflict Resolution

• Each Agent communicates with its neighbor(s)

• Tasks are allocated to the highest bidder based on the agent communication

• Lose the bid? Release task and all subsequent tasks

Re-execute if any task is not assigned an agent

Our Environment worked with : 3 agents and 6 tasksCopyright Prasanna Kolar, UTSA

Page 9: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 9

Traditional Intelligence Systems Vs Behavior Based Systems

Traditional Intelligence Systems (Non-Behavior based architectures)

• Functional Decomposition

• Sequential processing of functionalities (could be slow)

Traditional decomposition of a robot control system – functional modules

TASK EXECUTION!

Copyright Prasanna Kolar, UTSA

Page 10: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 10

Behavior Based System – Subsumption Architecture Example Layered Architecture – design, development starts at level 0

A parallel and distributed method for connecting sensors and actuators in robots

Behavior based robot control system

Copyright Prasanna Kolar, UTSA

Page 11: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 12

Subsumption – Steps for hardware implementation

UAV is instructed to take survey & take pictures of a blue object and send it back

Step 1: UAV takes offStep 2: Navigates towards targetStep2: UAV senses an obstacle in: this case a Red object

Starts the obstacle avoidance behavior and successfully avoids obstacleStep3 : UAV sensors do not detect an obstacle; UAV continuesStep4: UAV sensors detect wifi is down (UAV starts wandering)

CBBA is called and second UAV tracks the targetStep5: UAV senses; WIFI is back up; and follows This continues until the target is reached Step6: Take pictures of target and send back to base

Copyright Prasanna Kolar, UTSA

Page 12: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 13

Problem Statement Autonomous Surveillance with BCBBA

Ground Station Initiates Call to BCBBA

Tasks and the associated agents list is sent back to Ground Station

Subsumption Architecture module receives the list; starts the execution

Copyright Prasanna Kolar, UTSA

Page 13: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 14

Autonomous Surveillance .. Contd.

Task

s ex

ecu

ted

bas

ed o

n t

he

pri

ori

tiza

tio

n;

Sub

sum

pti

on

logi

c

Copyright Prasanna Kolar, UTSA

Page 14: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 15

Autonomous Surveillance .. Contd.

If more tasks; execute from beginning;Process is terminated when all tasks are completed

Copyright Prasanna Kolar, UTSA

Page 15: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 16

Simulation EnvironmentOperating system : Ubuntu 12.04Simulation software : ROS Hydro : Robot Operating SystemWith autonomous packages for AR Drone For GUI – Hector_quadrotorThis System simulates the UAV’s and their working environment.hector_quadrotor contains packages related to modeling, control and simulation of quadrotor systems

• Essential elements of ROS that we use:• ROS Packages : container that houses all the elements below• ROS Nodes : application codes that communicate and send

information• ROS Topics: Information channels that ‘carry’ the data• ROS msg: message containers/ ‘variables’ that help info sharing

between applications and also UAVs• ROS Publishers, Subscribers : these are the consumers and

broadcasters of informationCopyright Prasanna Kolar, UTSA

Page 16: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 17

Hardware EnvironmentProcessor and Memory Specifications:

1GHz 32 bit ARM Cortex A8 processor with 800MHz video DSP

TMS320DMC64x

Linux 2.6.32

1GB DDR2 RAM at 200MHz

USB 2.0 high speed for extensions

Wi-Fi b g n

3 axis gyroscope 2000°/second precision

3 axis accelerometer +-50mg precision

3 axis magnetometer 6° precision

Pressure sensor +/- 10 Pa precision

Ultrasound sensors for ground altitude measurement

60 FPS vertical QVGA camera for ground speed measurement

Video Specifications:

HD Camera. 720p 30FPS

Wide angle lens : 92° diagonal

H264 encoding base profile

Low latency streaming

Video storage on the fly with the remote device

JPEG photo

Video storage on the fly with Wi-Fi directly on your remote device or on a USB

key

Frame Specifications:

Carbon fiber tubes : Total weight 380g with outdoor hull, 420g with indoor hull

High grade 30% fiber charged nylon plastic parts

Foam to isolate the inertial center from the engines’ vibration

EPP hull injected by a sintered metal mold

Liquid Repellent Nano-Coating on ultrasound sensors

Page 17: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 18

Communication channels in ROS via topics• CBBA ‘node’ and subsumption

architecture ‘node’ are key here

• Nodes communicate with each other using ROS Topics

• CBBA node is started by the Ground Station;

• SA node subscribes to CBBA node through the integration system

• In the dynamic situation call, SA node places a call through integrated system to CBBA node during low battery, for a re-bid

• Similarly each drone can communicate via the non-node using topics

Copyright Prasanna Kolar, UTSA

Page 18: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 19

WiFi Down Simulation

Copyright Prasanna Kolar, UTSA

Page 19: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 20

WiFi Down Simulation

Copyright Prasanna Kolar, UTSA

Page 20: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 21

WiFi Down Simulation

Copyright Prasanna Kolar, UTSA

Page 21: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 22

WiFi Down Simulation

Copyright Prasanna Kolar, UTSA

Page 22: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 23

WiFi Down Simulation

Page 23: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 24

WiFi Down Simulation

Page 24: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 25

Simulation Results – Battery Down

Integration system formats info (SA consumes it)

Task Allocation system sends a stream of delimited text

Page 25: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 26

ASP Task Execution – Take off

• All 3 Drones Take off• Drone 3 is Lead Drone since it won Task• Drone 1 and 2 follow

Page 26: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 27

Drones Navigate to Target

• Drone 3 avoids an obstacle

• Drone 1 and 2 navigate and follow to target

Page 27: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 28

Drone 3 – battery low

• Drone 3 avoids obstacle 2 ; ‘Battery Sensor’ gives low battery

• CBBA is run and info subscribed

Page 28: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 29

Rebidding Call.. Drone 1 selection

Page 29: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 30

Drone 1 – New Task winner

• Drone 1 avoids an obstacle

• Drone 2 follows

Page 30: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 31

Drone 1 – low battery

• Sensors again give low battery on Drone 1

• CBBA is run and info subscribed

Page 31: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 32

Drone 1 – low battery – Rebidding call

• Re-bidding on BCBBA allocates task to Drone 2

Page 32: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 33

Drone 2 new winner – navigates to target

• Drone 2 successfully navigates to Target

• Drone 2 takes picture of Target & sends to base

• All Drones get back to base

Page 33: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 34

Drone 2 new winner - picture taken

• Drone 2 takes picture of Target & sends to base

Page 34: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 35

All Drones - return back to base

• All drones return to base

Page 35: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 36

Simulation Analysis

• Inter Module communication

• CBBA Bid call is processed without delay

• Automatic CBBA call for Rebid based on sensor readings

• Subsumption Architecture executes task as soon as the task is allocated

• Intra Module (Drones) communication

• Each drone is able to communicate with each other

Copyright Prasanna Kolar, UTSA

Page 36: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 37

System Analysis

Pros :

• The new system can exploit the distributed/decentralized nature of these 2 systems;

• CBBA guarantees convergence

• Parallel processing; quick reaction in Subsumption

Cons :

• No state retention in the traditional Subsumption Architecture

• Coding for memory in the programs will help

• Program designer should have a good system knowledge to program for the unforeseen anomalies

Page 37: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 38

BCBBA vs. CBBA

BCBBA Pros:Computational efficiency by about 20%Flexible : Leader drone can split tasks during execution

BCBBA Cons: Task split may require computation time

Copyright Prasanna Kolar, UTSA

Page 38: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 39

Conclusion

Objectives :

Develop an Unmanned Aerial System integrating a decentralized task allocation system using a Consensus based system& task execution system using Subsumption Architecture

Simulate and Implement de-centralized strategy for Surveillance with Dynamically changing Situations

Both these objectives have been accomplished in this thesis

Copyright Prasanna Kolar, UTSA

Page 39: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 40

Future work

• Current system did not have an offboard computer (like odroid), we used external computers for this processing.

• Use of systems like odroid xu4, raspberry pi – These can be integrated onto a more powerful drone.

• Eg: 3dr x8, iris, solo with odroid systems, raspberry systems.

• We have successfully integrated a 3dr pixhawk system with an odroid XU4 system, to enable complete autonomy.

• We had previously used default ROS messages

• Using broadcasting systems like LabStreamingLayer(LSL), we can communicate various different message types.

• We plan on building decentralized systems in the future using these technologies

Copyright Prasanna Kolar, UTSA

Page 40: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 41

References

1. Brooks, “A Robust Layered Control System for a Mobile Robot”, Robotics and Automation, IEEE Journal of; Mar 1986, pp. 14 – 23, vol. 2, issue 1

2. H.-L. C. a. J. P. H. Luc Brunet, "Consensus-Based Auction Approaches for Decentralized Task Assignment," in AIAA Guidance, Navigation and Control Conference and Exhibit, Honolulu, Hawaii, 2008

3. A. Whitten, H.-L. Choi, L. Johnson and J. How, "Decentralized task allocation with coupled constraints in complex missions," in American Control Conference (ACC), 2011.

4. M. A. a. J. P. How, "Robust Decentralized Task Assignment for Cooperative UAVs," in AIAA Guidance, Navigation, and Control Conference and Exhibit, Keystone, Colorado, 2006.

5. Wikipedia.org 6. ROS.org

Page 41: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 42

Questions ??

Please feel free to ask questions

Page 42: Implementing Consensus based tasks with autonomous agents ...techlav.ncat.edu/Presentations/Seminars-2016/Fall... · Wide angle lens : 92° diagonal H264 encoding base profile Low

The University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 782499/2/2016 43