3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
Brigham Young University Indoor MAV System P. Travis Millet* and Neil Johnson† Brigham Young University, Provo, Utah, 84602 and Gregory M. Alldredge‡ Brigham Young University, Provo, Utah, 84602
Brigham Young University has developed a Vertical Take-Off and Landing (VTOL) Micro Air Vehicle (MAV) capable of autonomous flight in GPS denied environments and capable of fulfilling the objectives of the 3rd U.S.-European indoor competition for Micro Air Vehicles. The BYU indoor MAV system is a coaxial helicopter that weighs less than 480 grams and measures 450 mm from rotor tip to rotor tip. The system is equipped with a Kestrel Autopilot made by Procerus Technologies which uses proportional-integral-derivative control to realize autonomy. In order to navigate the course without GPS data the MAV relies on information about its position gained through a micro CCD camera and from the sensors integrated into the Kestrel autopilot. An optical flow algorithm is used to determine location and a feature tracking algorithm is used to navigate through a window and for precision landings. The MAV is also equipped with both hardware and software safety features in order to ensure a safe flight.
*
Graduate Research Assistant, Multiple Agent Intelligent Cooperative Control (MAGICC) Laboratory, Brigham Young University, 966 Eastgate Drive, Provo, UT 84606. † Graduate Research Assistant, MAGICC Laboratory, Brigham Young University, 951 East 150 North, Provo, UT, 84601. ‡ Undergraduate Research Assistant, MAGICC Laboratory, Brigham Young University, 735 North 400 East #25, Provo, UT 84606. 1
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
I.
Introduction
A
s technology has progressed and the hardware required for avionics packages has decreased in size, the ability to realize fully autonomous flight on micro air vehicles has come to fruition. Brigham Young University has made significant contributions to the field of fixed wing micro air systems by developing a small, lightweight autopilot system which is now being adapted for achieving autonomous flight on a rotary wing system to be used in the 3rd US-European Competition and Workshop on Micro Air Vehicles (MAV07). The system that Brigham Young University students have developed for the MAV07 Indoor Competition is an electric powered, coaxial helicopter weighing less than 480 grams and measuring 450mm from rotor tip to rotor tip. Because the competition will be held indoors the coaxial helicopter cannot rely on GPS information in order to navigate itself. Due to electrical wiring and metal obstacles that may influence magnetic fields use of an onboard magnetometer has also proven to be difficult. This means that the indoor system needs to be able to navigate via the gyros, accelerometers, and vision sensors that it carries onboard. Autonomous navigation by means of these methods exhibit inherent difficulties and the BYU team therefore is using a combination of several of these methods in order to achieve the objectives of the MAV07 indoor competition. The MAS is equipped with a Kestrel v2.2 autopilot manufactured by Procerus Technologies which uses integrated solid state rate gyros and accelerometers to estimate roll, pitch, and yaw angles and rate changes. With this information the autopilot uses proportional-integral-derivative control method techniques in order to achieve autonomous flight. The autopilot communicates to a ground station via a 2.4GHz modem signal and relays telemetry, attitude estimation, and video back down to the user at the groundstation where much of the navigation processing takes place. The rotary wing system is also equipped with both hardware and software failsafes in case of situations such as loss of communication with the ground station or erratic flight behavior in order to ensure the safety of persons, property, and the aircraft itself. The BYU Multi-Agent Intelligent Coordination and Control Laboratory (MAGICC Lab) GPS denied rotary wing MAV system is a collection of proven and newly developed technologies fused into a platform that is robust, safe, and capable of handling operator-defined objectives.
Figure 1. Figure 1 shows the BYU Rotary Wing MAV System next to a 12 inch ruler.
2
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
II.
Airframe
The BYU MAV system airframe is built upon of a modified Walkera Z-53#1 coaxial RC helicopter (see Figure 1). All extra parts that amounted to excess weight were removed from the stock RC helicopter, this included most of the skeleton that formed the support of the aircraft. This support was restored to the helicopter by means of G10 glass reinforced composite plates that now constitute the structural rigidity and body of the vehicle. The original motor bracket proved to be excessively flimsy and was therefore replaced by a significantly thicker custom aluminum bracket. The 400 series brushed motors were replaced with RZ-Micro Heli v2 brushless motors due to their significant weight reduction and increased power. Each motor is regulated by a 25 Amp Phoenix Speed Control made by Castle Creations. All electronics except for the servos we taken off the aircraft as their functions were replaced by the autopilot and data link system. A small ring of foam was attached to the bottom side of the aircraft in the event of hard landings and to prevent rotor strikes.
Figure 2. Figure 2 shows a stock Walkera Z-53#1 RC helicopter before any of the unnecessary parts were removed or replaced.
III.
Computer Vision Hardware
The helicopter is equipped with a KX-141 Color CCD camera from Black Widow Audiovisual1 which provides 30 frames per second at 640x480 pixels per frame. The color CCD camera is fitted with a 69 degree field of view lens in order to provide the system enough detail for vision based navigation but also allowing room for to vibration or drift in the video while still maintaining all points of interest. The camera is mounted on a 1-axis gimbal (see Fig. 3c) which allows the user to control the azimuth of the camera over a range of 130 degrees. The system relays video to the ground station via a 50mW micro video transmitter (see Fig. 3a) which transmits at 2.4 GHz. This transmitted signal is received at the ground station via a 200mW video receiver (see Fig. 3b) and is then passed through a K-World frame grabber, which digitizes the analog video stream for further processing at the ground station.
a. b. c. Figure 3. Figure 3a shows the 50mW micro video transmitter used on the BYU MAV System, Figure 3b shows the 200mW video receiver used to receive video from the MAV, Figure 3c shows the 1-axis gimbal which allows the camera to have a variable elevation angle.
3
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
IV.
Sensors
The conventional method of determining altitude for aircrafts is by measuring barometric pressure and comparing this to a pressure reading taken on the ground. This becomes a more difficult problem in rotary wing aircrafts because rotor wash will likely degrade the barometric pressure reading. Indoor flight applications also require a greater resolution of altitude than barometric pressure readings can provide. For this reason the BYU rotary wing MAV uses a high performance sonar range finder for its height above ground measurements. The system uses a LV-MaxSonar EZ1 ultrasonic transducer2 (see Fig. 4a) which is able to detect objects from 0 to 6.45 meters with 2.5 centimeter resolution. The ultrasonic transducer on the BYU aircraft points straight down for height above ground measurements though it could be used for other applications such as obstacle avoidance or 3D mapping as well. The BYU team also experimented with using a MicroMag 3-Axis magnetometer (see Fig. 4b) for navigation purposes but found it difficult to get it far enough away from the magnetically driven motors on such a small airframe to avoid magnetic interference.
a. b. Figure 4. Figure 4a shows the LV-MaxSonar EZ1 ultrasonic transducer used for altitude measurements on the MAV, Figure 4b shows the MicroMag 3-Axis magnetometer that proved too unreliable to be used on the final system.
V.
Data Link
The aircraft communicates wirelessly via an onboard 2.4 GHz XBee Pro Modem (see Fig. 5) made by MaxStream3 which provides communication for up to 1.6 kilometers. The ground station is equipped with a Maxstream commbox that has been modified and reprogrammed to communicate with the XBee Pro line of modems.
Figure 5. Figure 5 shows the 2.4 GHz XBee Pro Modem that is used on the BYU MAV System.
VI.
Autopilot
The MAS carries a Kestrel 2.2 autopilot (see Fig. 6) manufactured by Procerus Technologies4. The autopilot is renowned as one of the smallest commercial autopilots in the world weighing in at just 16.65 grams and with the petite dimensions of 51mm x 35mm x 12mm. The autopilot is integrated with 3-axis rate gyros and accelerometers for attitude estimation. The autopilot uses a Rabbit RCM 3400 microprocessor to perform low level navigation and stabilization tasks as well as monitoring communication with the ground station. The autopilot is traditionally preprogrammed to fly fixed wing aircrafts and communicates with the ground station via a software program called Virtual Cockpit ™. In order to fit the needs of a rotary wing aircraft the autopilot had to be reprogrammed with different methods for attitude estimation, navigation, and stability control. Likewise the software that the ground station uses to command the fixed wing aircrafts aircraft had to be revamped to fit the needs of a rotary wing aircraft (see Fig. 7).
4
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
Figure 6. This figure shows the Kestrel 2.2 autopilot in comparison to a quarter.
VII.
Actuators
The BYU coaxial helicopter uses a method known as cyclic pitch and cyclic roll in order to produce pitch and roll angles. A pitch or roll angle is produced by changing the feathering angle (angle of attack) on the lower blades at a certain phase of their 360 degree rotation, thus making the feathering angle a function of the azimuthal position of the rotor blade. Coaxial helicopters also have a unique method of yaw control. Since there are two sets of blades and they rotate opposite of each other yaw control can be achieved by running one set of blades faster then the other and therefore creating a greater yaw moment in one direction. Yaw in the opposite direction is achieved by the same idea. Coaxial helicopters that use this form of yaw control have no need for tail booms or tail.
VIII.
Control Methodology
The commercial autopilot as well as the revamped model used for the rotary wing craft both use proportional-integral-derivative (PID) control with successive loop closure to achieve autonomy. PID control is capable of providing very tight and accurate control of the aircraft. Much of the ground station software helps the user to tune the PID gains in order to reduce overshoot while still attaining desired values. While the end goal is for the aircraft to be flown under full autonomy of the autopilot the user can take over at any time and with any varying degree of computer help. Due to high frequency vibration on the aircraft the output of the attitude estimation software did not provide accurate approximations for the aircrafts attitude with estimation error as high as 40 degrees in pitch and roll. This problem was resolved by estimating pitch and roll by passing the raw gyro and accelerometer data through an alpha filter and then passing the filtered data into a Kalman Filter5.
Figure 7. This figure shows the map screen and artificial horizon screen of the ground station software. 5
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
IX.
Dead Reckoning
On fixed wing aircrafts velocity is easily estimated with the use of a pitot sensor or by comparing previous and current GPS data. Much like the problem of barometric pressure readings on rotary wing aircrafts, the pitot sensor readings become highly inaccurate in the rotor wash of the helicopter. Indoor navigation also requires more precise estimations than a pitot system can provide. For this reason other options are explored in order to solve the problem of knowing where the aircraft is and the aircrafts is moving. One method of determining x, y, and z velocities is via a method called dead reckoning. This is done by integrating the gyros and accelerometers over a period of time. Thereby knowing the accelerations from the accelerometers and the angular rates from the rate gyros, position relative to where the aircraft started should be able to be determined as a function of time. However, this method has proven to have inherent problems. The rate gyros on the autopilot are especially prone to drift. This fact coupled with the fact that a small error accumulates rapidly when integrated over time induces considerable error in position estimates. For this reason dead reckoning can only be trusted for a few seconds before enough error has accumulated to make it unreliable.
X.
Vision Based Navigation
Due to the drawbacks of navigating by dead reckoning the BYU helicopter also is capable of navigating using some basic vision based algorithms. Pose estimation is a method of determining roll, pitch, yaw, and position relative to an object based on the known dimensions of an object. Of course this does not work in all scenarios especially when exact dimensions of objects are not known. In the case of the indoor competition, where we know several dimensions of objects (i.e. the window, the three targets, the pedestal, etc.), this method gives very accurate pose and position estimation. The pose estimation algorithm6 that was written by BYU researchers and used on the helicopter is accurate to within 15cm at a length of 5 meters away from the known object with similar accuracy in estimating pitch, roll and yaw. For an example of the BYU pose estimation application see Figures 8 and 9. The only disadvantage to pose estimation is that the known object must always be in the video screen in order to know attitude and position. For such a small helicopter in such close quarters, keeping something in view can be a somewhat difficult task.
Figure 8. This figure along with Figure 9 shows the pose estimation program that the BYU MAV System uses to determine position relative to the center of the object, the center of the object is marked by the green dot. The points that the blue dots are tracking form a 1 meter by 1 meter square.
6
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
Figure 9. This figure along with Figure 8 show the pose estimation program used for position estimation of the BYU MAV. The MAV was two meters away from the ‘window’ and it can be seen in the z translation box under ‘Pose Estimation’ that there is an error of only 8mm. Another vision based means of calculating position is based on the optical flow7 of features in the video. The simplest case of this is when the camera is pointed straight down it can see several features and using a feature tracking algorithm the aircraft can calculate how those features move in time. Upon the assumption that nothing is moving except for the aircraft itself velocity and position can be found based on how things have moved underneath the aircraft. This method has not yet proven to yield very accurate results due to the need of further filtering of the data to track features. Another problem arises in the fact that the helicopter has pendulum-like stabilization and is therefore liable to sway back and forth outputting velocities which actually don’t exist.
Figure 10. This figure shows the optic flow estimation program that the BYU MAV System uses to estimate x and y velocities relative to its body frame with a downward facing camera, the red lines are projected velocity vectors.
7
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
XI.
Safety Procedures and Failsafes
Safety is a vital part of any system especially when a malfunction of the system could result in the injury of persons or the destruction of expensive equipment. Recognizing the need for safety precautions, the BYU MAGICC Lab rotary wing MAV system has the ability to manually override the autopilot and has built in failsafes in the case of system failures. A preflight checklist is also followed to ensure proper performance of the system before the MAV is put in the air. Fail-safes are a system of preset instructions used to control the MAV in the event of a subsystem failure. The BYU rotary wing aircraft is equipped with both software and hardware fail-safes in the unpredicted or erratic flight behavior. During flight the aircraft is both sending and receiving information from the ground station. In the event that the helicopter doesn’t receive any information from the ground station for some predetermined time (default is 3 seconds) then the helicopter performs an autonomous landing but keeps the motors running. In the event that the helicopter still hasn’t heard from the ground station after a cumulative predetermined amount of time (default is 7 seconds), then the motors shut off. At any time during communicating flight the safety pilot can take control of the helicopter by means of a radio controlled (RC) transmitter to safely land or navigate the vehicle out of danger. The RC transmitter is also equipped with a kill switch which simply turns both of the motors off instantaneously.
8
3rd US-European Competition and Workshop on Micro Air Vehicle Systems (MAV07) & European Micro Air Vehicle Conference and Flight Competition (EMAV2007), 17-21 September 2007, Toulouse, France
References 1
“Black widow av,” http://www.blackwidowav.com, [cited Sept. 4, 2007]. “Maxbotix® inc.,” 2006, http://www.maxbotix.com, [cited Sept. 4, 2007]. 3 “Maxstream,” http://www.maxstream.net/, [cited Sept. 4, 2007]. 4 “Procerus technologies,” http://www.procerusuav.com/, [cited Sept. 4, 2007]. 5 Beard, R. W., “Guidance and Control of Autonomous Miniature Air Vehicles,” Brigham Young University (to be published). 6 Ready, B. B., and Taylor, C. N., “Visual Odometry by Direct Registration for MAV Navigation in GPS-Denied Environments,” Brigham Young University (to be published). 7 Stavens, D., “Introduction to OpenCV,” http://ai.stanford.edu/~dstavens/cs223b/stavens_opencv_optical_flow.pdf, [cited Sept 4. 2007]. 2
9
