Simulation and Experimental Validation of Task Allocation Algorithms for Uavs

PROJECT SUMMARY REPORT

SIMULATION AND EXPERIMENTAL VALIDATION OF TASK ALLOCATION ALGORITHMS FOR UAVS

Submitted To

The 2012 Academic Year NSF AY-REU Program

Part of

NSF Type 1 STEP Grant

Sponsored By

The National Science Foundation

Grant ID No.: DUE-0756921

College of Engineering and Applied Science

University of Cincinnati

Cincinnati, Ohio

Prepared By

Tim Arnett, Junior, Aerospace Engineering

Devon Riddle, Junior, Aerospace Engineering

Report Reviewed By:

Dr. Kelly Cohen

Dr. Chelsea Sabo

University of Cincinnati

September 17 – December 6, 2012

CEAS REU Project #4: Experimental Testing of Allocation of Multiple UAV’s

Goals and Objectives: The point of this research was to verify a task allocation algorithm with experimental data and througha simulation program. The overall two goals of this research were to learn to interface equipment and to compare two routing solutions for common performance metrics. These two goals were defined in a set of three objectives. The first objective was to interface with cooperative control development systems. By this we interfaced and ran algorithms on AR Drones as well as through AMASE, the software simulation program. This was done to gather the needed experimental data to test the test allocation ofUAV’s in both an experimental and simulation test environment. Next the task allocation algorithm that was developed was validated both experimentally and through simulation. This algorithm was written with the comparison between distance travelled and the delivery time for time critical targets in mind. Lastly, the cooperative control strategies for the UAV’s were tested and compared. These strategies included the totaldistance a UAV travelled and the total delivery time cost for time critical targets.

Research Tasks:Below is a set of research tasks that were used to accomplish the goals and objectives that were mentioned above.

1. Install necessary hardware.
2. Students become familiar with the software through personal training.
3. Learn to interface with cooperative control development systems
4. Working simulations are developed.
5. Implement code on physical systems.
6. Implement cooperative control algorithm in MATLAB.
7. The experimental data gathered is transcribed into the simulation software.
8. Results are gathered for experiments performed with AR Drones.
9. Results are gathered for simulations.
10. Solutions gathered experimentally and through simulation are tested and compared.

Methods:This algorithm was the waypoint algorithm and utilized control methods that included PID control and fuzzy logic control. The PID control was used for the yaw and ascent rate of the rotorcraft. The benefits of using the PID control was a good response and settling time was provided and the simple implementation. The Fuzzy Logic controller was used for pitch and roll and it did not require a system model and was robust to stability issues.

Having the waypoint algorithm,a minimum distance solution and a minimum delivery latency route were developed. The minimum distance route is useful for minimizing the total mission time, fuel consumption, and overall energy that the UAV puts out during the mission. Unlike the minimum distance route, the minimum delivery latency route is desired when data needs to be delivered to a high bandwidth connection or a “depot.” Once both routes were developed, they were applied to tests that were performed both experimentally in an IMAGE lab with the AR Drone UAV’s and in simulation on a computer program environment known as AMASE. AMASE is an Air Force flight simulation environment that was already proved as a legitimate way to set up realistic flight simulations. In each environment, three different tests were performed that varied in difficulty and the number of targets and the minimum distance and minimum delivery latency solutions were implemented for each test. When the three tests were concluded, the distance travelled and the delivery times for each test were compared.

Results: For the implementation on the actual robots, different parameters were used due to scaling issues. The resulting formation approached a state that was relatively stable. Issues arose where the robots would oscillate in formation. This can be attributed to a few possible causes. Actuation errors between the commanded velocity for the robots and the actual velocity, image noise from the cameras, or update timing issues. This can be refined in future work.

Taking the data that was gathered through MATLAB, the points were converted into latitude and longitude coordinates so they could be implemented into AMASE. The point of transcribing the data into AMASE was to prove the validity of the task allocation algorithm previously developed. Through AMASE, it was found that the simulated results were similar to that of the results gathered experimentally. However, in one or two of the cases there were a few errors that stood out. This could have been due to the bank angle that AMASE has a requirement of when simulating a scenario or it could be a software error. AMASE was not made to simulate scenarios where the distance between point A and point B was so small. As a result, the points were up-scaled; however, the distance relationships were identical.

For the analysis between the results gathered experimentally and through simulation, the total time cost for the minimum delivery latency solution was found to be 751.96 seconds while for the minimum distance it was 1134.92 seconds with a difference of about 34%. For the total distance travelled, the minimum delivery latency solution was calculated at 25.11 seconds while the minimum distance solution was calculated at 18.04 seconds with a difference of about 28%. Between the minimum delivery latency solution and the minimum distance solution, there was a different of about 13% when it came to the total time cost.

Training Received: The majority of the training focus was on the use and compatibility of the software applications that were needed for this project. Training was necessary for the following software: OptiTrack Tracking Tools, and AMASE.

Training was also provided for the camera systems. Calibration was necessary to ensure accurate data acquisition. The Optitrack system, which uses multiple cameras to track objects in three dimensions, has to be calibrated with a reflective wand. Although this system was not used for the ground robots in this project as they only move in two dimensions, it will be used for future work involving unmanned aerial vehicles (UAVs). The overhead camera system, used for the two dimensional images requires distinguishable patterns to be attached to the top surfaces of the robots.

The last computer program that involved training was AMASE. AMASE was given to us by the Air Force Research Laboratory at Wright Patterson. AMASE is a desktop simulation environment that was developed for UAV cooperative control studies. It was already proven as a legitimate way to set up realistic flight simulations and the only help that was received in learning about the simulation software was a vague tutorial that came with it. Beyond the tutorial, the training was done by playing around with AMASE and figuring out what worked with what and what parts of the simulation went where by trial and error. Once the basics of AMASE were figured out, simple simulations could be created in the environment that would mirror what was done experimentally.

Conclusion: In conclusion, the experimental tests confirmed the results for both total time cost and total distance travelled for the two different routing paths. The difference in the values for the minimum distance for the ideal and the tests could be attributed to errors from the camera system and the robustness of the waypoint algorithm. For the delivery latency tests, it is important to note that even with these errors, the total time cost was still much lower for the minimum distance case. While it may seem that this is expected, the fact that the UAV has a larger distance travelled in the case of the minimum delivery latency route, it shows that even with more chance of path deviations, the total time cost is still much lower.

In addition, AMASE proved to be a capable simulation environment save for the few problems that were experienced when trying to task the UAV with following a specific path line. On more than one occasions, Stephanie Lee, a graduate student at Wright Patterson who helped to develop AMASE, was called upon to help figure out a few errors that the tutorial did not mention. It was discovered in the end that AMASE was made to simulate tasks for UAV’s flying a few kilometers apart. This was a slight issue, since to make the simulation as realistic as possible; we wanted the distances between point A and point B to be about a few meters. A simple solution was found in up scaling the waypoints given to the UAV and the basic percentages were compared.

The future work in this area could involve implementation on a larger scale for multiple UAV’s. There are plans to construct and fly larger and more powerful quadrotor UAV’s in order to perform tests outdoors. There will also be more waypoint algorithm development and refinement in order to more closely follow the ideal path. These steps will also provide more validation before full scale flight tests by the USAF.

Photographs:Figure one below showcases two action pictures for this research project. Picture (a) was taken from a bird’s eye view of where the rotorcraft is being brought back to its start point before setting it on its path. Tim is placing the rotorcraft back at the starting position before running another experimental test. In picture (b), research student Devon is working with the simulation environment, AMASE, in the development of one of the scenarios provided through experimental data.

(a)

(b)