Missouri S&T AESS UAV Team

On May 10th we attempted our first bottle drop from 400 feet.  The bottle had an aluminum tab attached to it which was held by small servo at the center of gravity of the UAV.  We took off under manual control and gained altitude, as the UAV flew it’s oval path autonomously, the bottle was released over the field.  The first bottle we dropped was a 500mL Nalgene bottle, which failed miserably (see picture below).

With the water bottle attached, the flight characteristics were not affected, although takeoff distance was increased, the autopilot’s ability to control the UAV was not affected noticably.  During these flights the winds were relatively high, ranging from 10 to 15 MPH, which made navigation for the UAV difficult, but the autopilot performed admirably.  Though at times it was blown off the path, it returned quickly to the intended flight path.  Several autonomous test flights were performed in the windy conditions, further demonstrating the robustness of our UAV platform.

Three more test flights were performed on May 15th, all of which included water bottle drops with different bottle designs that took into account lessons learned from the first test.  Enclosures were developed for subsequent bottles that allowed them to survive the 400-foot drop.  Data was also gathered about where the bottles landed and at what point they were released.  With this data the theoretical and actual horizontal distance traveled were compared.  After analysis it turned out that the horizontal distance traveled from two of the drops was consistent, this allowed a simple model to be developed for the trajectory of the bottle.  During the final flight, aerial video of the bottle being released was also obtained from the onboard digital camera (see the video below).

These test flights provided valuable data about how the bottle’s trajectory is affected by the wind resistance, as well as about how the UAV platform handles moderate winds.  The next crucial step will be to integrate the onboard computer along with the camera into the aircraft to allow testing of the system as a whole.

On Saturday (April 5th) we flew the airplane for the first time with the autopilot since it got the new wing.  But before we went out to the RC airfield we had to physically install the autopilot in the fuselage, which went well.  We then performed all of the flights necessary to tune the autopilot in HIL (Hardware In the Loop) simulation mode.

Using HIL simulation allowed us to all to get a better of idea of what we had to do once we got out to the airfield.  In the end performing the HIL simulation in the lab saved us a lot of time and allowed me to get more familiar with tuning the PID control loops for the autopilot.  Previously we had not used the PID window in Virtual Cockpit, but after using it in the HIL simulation it proved to be invaluable for properly tuning the autopilot, it allows you to see the actual, desired, and effort of a certain control parameter (i.e. pitch, roll, yaw, altitude, etc.).  Previously we had had issues seeing a change in the behavior from the ground after changing a certain PID gain.

On the first flight soon after takeoff the aircraft became unstable and began to become uncontrollable and began oscillating wildly, fortunately our pilot Kyle was able to get it back on the ground without any damage.  It turns out that our CG was too far aft, but after adjusting the CG we were able to continue with testing.  Using the PID window and the having performed the same process of tuning the autopilot in HIL mode, the first few flights went quickly and we were able to quickly tune the level 1 control loops within two 15-minute flights.  A graph of the autopilot’s roll performance is shown below, ideally the two lines should match, and they are in fact very close.

We then moved on to the level 2 control loops (i.e. pitch from airspeed, pitch from altitude, airspeed from throttle, etc.).  These were slightly more difficult, and took several passes over the airfield to complete, but we were able to get them tuned to a reasonable level.  The ability of the autopilot to maintain a constant altitude is shown below, there are slight oscillations in the altitude (+/- 3 meters), these are reasonable, but might be improved at a later time with more tuning.

Just as my laptop battery was about to give out we moved on to the final flight of the day.  The purpose of the final flight was to verify that the autopilot is able to navigate accurately and safely.  The first test was simply placing a loiter waypoint above the center of the RC airfield; although there were some oscillations in altitude, the overall performance was quite good (note the 4 m/s wind speed).

The final test was to create an oval over the airfield that the UAV would have to navigate.  The first time around the circuit the UAV did not adhere to the waypoints too strictly, but after adjusting some of the navigation parameters the performance was significantly improved (see image below).  After completing these flights the telemetry was reviewed and we were better able to analyze the performance of the autopilot and its navigation.

Overall the testing went well and we were able to tune the autopilot control loops better than we had previously been able to with the old wing, this is most likely due to our use the PID window and the HIL simulation which allowed our time at the airfield to be spent much more efficiently.

And now for some video of the flight and the telemetry from the navigation flights:

I recently started assembling a test rig to integrate all of the electronics that will be going in the UAV. By performing all of this integration on test bench, the final installation and configuration will be performed much faster. This will also allow us do our HIL (Hardware In the Loop) simulation with all of the components connected.

You may have noticed that there are two large NiHM battery packs, these will soon be replaced by two (11.1V 3200mAh) Lithium Polymer battery packs, saving us about half a pound of weight and significantly increasing our power capacities. Although for bench testing all of the electronics will be powered of off a standard ATX computer power supply. The next step is the get the HIL simulation working with the autopilot and the it’s simulator, this part has been a bit finicky but it should be resolved now.

As a team we have also discussed possible strategies for image acquisition and processing. We had previously planned on simply acquiring VGA resolution video at 30 frames per second, but after reconsidering what altitudes we will be required to fly at to cover the search area in a reasonable time we quickly realized that VGA video would not provide the resolution we need to identify a human target on the ground from an altitude of 400 feet. We have since decided it would be best to use a much higher resolution still camera taking images at set intervals and tagging them with the GPS coordinates and the orientation of the UAV.

We went out flying a few weeks ago (September 8th) with the intent of tuning the autopilot control loops better, since in their current state they were not tuned well enough for autonomous landing. We headed out to the RC field around 3 and shortly got everything together. We had also installed a camera on the vertical stabilizer to finally get some in flight video. For the most part everything was going smoothly until the third flight.

During the third flight shortly after takeoff we began a descent and then both wings suddenly tore off from the fuselage. It is important to note that this occurred under the manual control of the pilot, it should also be noted that the UAV was carrying more weight than we have had in any other flights. We had also previously suspected that the wings might be too weak to handle the weight, and it turns out they were. After the wings tore off, the airplane was still several hundred feet in the air and it quickly went into a dive and plummeted into an open field, thankfully no cows were harmed in the crash. Below are pictures of the carnage…

As you can see from the pictures above the fuselage is completely ruined, as are the wings, but thankfully all of the electronics and the engine survived. We have already begun to search for a new airframe, currently we have not been able to find an aircraft that is more suited for our purpose so we will most likely be purchasing another Sig Kadet and building our own wings in order to make them much stronger and more efficient (by using a different airfoil).

Here is some in-flight video before the airplane crashed…

Click here to download the high quality video.

We had our first few autonomous flights two weeks ago, and the UAV didn’t crash, unlike another airplane that day. We got out to the RC field in the early afternoon and it started to rain, it rained on and off all afternoon, so we were only able to fly for short periods of time in between the rain. There were also several other planes out at the field that day, including one of the senior project airplanes built entirely out of carbon fiber.

Unfortunately that airplane was extremely hard to control and on it’s first flight it crashed shortly after takeoff, shearing some nylon bolts. After replacements arrived, it took off again and the second flight lasted much longer, approximately five minutes, before the pilot lost control and it plummeted into the ground. Just as a note, the pilot was a quite experienced, and still had quite a bit of trouble controlling it. Now for the carnage…

As you can see, not much was left of that…

Thankfully our flights did not end like that. Once the runways were clear we prepared for our first autonomous navigation with the autopilot. I created a simple path in the shape of a rectangle for the UAV to fly over the RC field. The UAV took off under manual control and once it was in the air control was handed over to the autopilot. The first time it flew in autonomous navigation mode, the path that the UAV took was rather sloppy, and not very close to the waypoints that were specified. After giving a quick call to Procerus and adjusting a few parameters it flew much better and closer to the path designated. It was quite the sight seeing the UAV fly itself, a little scary at times, but also very impressive.

After flying a few simple paths and loiters, we attempted to use the autonomous landing mode of the Kestrel autopilot. The UAV began to circle down to the proper altitude and once it reached the altitude where it was supposed to come in for the landing approach, for lack of better words, it simply wandered off into the distance and we were forced to take manual control and bring it back. After a few attempts at it, all resulting in the same behavior, we decided to give up and just send the telemtry log back to Procerus to see if they can diagnose the problem.

In this picture you can see the pitot tube to measure airspeed on the wing.

That weekend we went out flying again on Sunday and attempted our first autonomous takeoff, which worked quite well the first time. Once again we attempted some autonomous navigation, this time going farther from the home location (over 1.5 kilometers or ~1 mile). Several times during those flights we had to bring the UAV back under manual control since there was real air traffic to the nearby airport and we did not want to interfere. I will be posting video of our first autonomous flight soon, as well as an update on our electrical system soon.

We had our first flight of the airplane with the autopilot installed last Tuesday (April 17th). We arrived at the RC airfield around 6PM, much to our dismay there was some light rain, which eventually stopped. Setting up the ground station was the first priority, then all of the control surfaces were tested to make sure they were functioning and deflecting the right amount, although for some reason the throttle servo was not moving, after a little bit of tweaking in the Virtual Cockpit, it worked fine. Shortly after the throttle servo was fixed, the autopilot stopped responding, it turned out that the ground connection of the battery to the autopilot had failed, since we did not have a soldering iron handy, the two wires were twisted together and wrapped securely in electrical tape, although as an interesting note, if this wire would have come lose in flight, all control of the aircraft would have been lost.

Abe and Jonathan starting the airplane

The moment of truth came and we fired up the engine and began the to takeoff (on manual control), the airplane was able to takeoff in a relatively short distance and was quickly in the air. The radio link between the autopilot and the ground station was quite strong and the Virtual Cockpit provided us with the airspeed, height above ground, and the attitude of the aircraft. The first order of business was to calibrate the dynamic and static air pressure sensors on the airplane. The next few flights consisted of tuning the control loops with the correct parameters, this was fairly straightforward, but rather time consuming. Once the roll and yaw loops were configured, it was interesting to see how the airplane quickly straightened itself when the pilot attempted to roll it. Although we were not able to finish tuning the control loops since it began getting dark.

Overall the first flight with the autopilot was a great success. We plan on flying the airplane again this Sunday, and will hopefully finish tunning the control loops and have our first autonomous flight.  For more pictures you can visit our flickr site.

This weekend we began installing the Kestrel autopilot in the airplane, which consisted of wiring up all of the servos and GPS, as well as configuring the servos in the Virtual Cockpit software. This took a surprisingly long time, although most of the time was spent on making the wiring harnesses and troubleshooting why the servos were not function properly. Once everything was powered up, the servos didn’t want to move, which it turns out was because I did not use the autopilot’s power rails since I had completely bypassed them and sent only the PWM signal wire from the autopilot to the servos, and powered the servos independent of the autopilot. Once that autopilot’s power rails were hooked up the servos started working, although they were quite jittery and very rough when on manual control, possibly from some kind of interference from the radio or other electronics. Below are some pictures of the autopilot installed in the airplane.

The autopilot

The GPS unit.

Now that the autopilot is installed, and as long as there is good weather this week, we should be able to have the first few flights with the autopilot. According to the manual, it will only take about four flights to tune the control loops for fully autonomous flight. We will also probably end up purchasing an embedded computer this week for the image processing, which will allow us to start developing the image processing algorithm.

Last week the the Kestrel Autopilot from Procerus arrived, and I have to admit, I was quite surprised by the size of it. It’s literally smaller than a deck of cards (slightly taler but less area). Reading over the documentation was first step in learning how to use this powerful little thing. Thankfully the documentation was quite good, and after 5 minutes of soldering a few wires together, it was ready to be powered on for the first time. The accompanying software was quite straightforward and surprisingly once both the autopilot and ground station unit were powered, they instantly acquired a wireless link, and began feeding data into the Virtual Cockpit software.

One really cool and amusing feature was that there was an artificial horizon with a 3D model of an airplane that rotated in whichever direction the autopilot was oriented. After everything was configured roughly how it should be, I began to setup the HIL (Hardware In the Loop) simulation. This allows the autopilot to fly an airplane in a simulator on the computer, allowing the simulation of an entire flight from takeoff to landing. The HIL simulation was quite easy to setup, and was running within minutes. Of course there were several minor details that were overlooked, due to the fact that I had not read the entire manual yet and did not understand the various modes of the autopilot, so the UAV kept crashing, but eventually after playing around with it, it all worked great.

Overall I was rather impressed with how straightforward and easy it was to start using the autopilot right out of the box. From my experience I’ve rarely had such complicated systems up and running in such a short amount of time. Now the next step is to install the autopilot in the airplane, which is quite a task in itself. Hopefully we will be able to have it installed by the end of this weekend, and maybe perform the first flight with the autopilot this weekend, or sometime next week, weather permitting.

And now for some pictures:

The autopilot itself.

The communications box.