J Intell Robot Syst DOI 10.1007/s10846-011-9569-1
A Data Fusion System for Attitude Estimation of Low-cost Miniature UAVs Long Di · Tobias Fromm · YangQuan Chen
Received: 29 January 2011 / Accepted: 13 April 2011 © Springer Science+Business Media B.V. 2011
Abstract Miniature unmanned aerial vehicles (UAVs) have attracted wide interest from researchers and developers because of their broad applications. In order to make a miniature UAV platform popular for civilian applications, one critical concern is the overall cost. However, lower cost generally means lower navigational accuracy and insufficient flight control performance, mainly due to the low graded avionics on the UAV. This paper introduces a data fusion system based on several low-priced sensors to improve the attitude estimation of a low-cost miniature fixed-wing UAV platform. The characteristics of each sensor and the calculation of attitude angles are carefully studied. The algorithms and implementation of the fusion system are described and explained in details. Ground test results with three sensor fusions are compared and analyzed, and flight test comparison results with two sensor fusions are also presented.
L. Di · T. Fromm · Y. Q. Chen (B) Center for Self-Organizing and Intelligent Systems (CSOIS), Department of Electrical and Computer Engineering, Utah State University, Logan, UT, USA e-mail:
[email protected] L. Di e-mail:
[email protected] T. Fromm e-mail:
[email protected]
Keywords Data fusion · Multi-sensor fusion · UAV navigation · Inertia measurement · Estimation · Parley
1 Introduction The development of miniature UAVs has achieved tremendous progress in the last several years with improvements on avionics, airframes and payloads, especially regarding size, material and power consumption [1, 2]. Miniature UAVs have attracted wide interest because of their numerous applications in civilian, agricultural and military areas. Since non-military UAV developments put more emphasis on the low-cost feature [1], navigation systems based on low-cost sensors become an appropriate solution to satisfy this attribute. Low-cost sensors can provide measurements with relatively lower accuracy compared to expensive industrial or commercial sensing systems. This is usually because they are restricted by employing less accurate hardware and less sophisticated software. Due to the constraints on the precision of low-cost sensors, they cannot be fully relied on to play an important role in application such as attitude estimations of aircrafts. But if there are several low-cost sensors, each with different characteristics, it is possible to achieve improved accuracy by using genuinely designed
J Intell Robot Syst
fusion algorithms. In this case, each of the lowcost sensors can be used to measure the attitude of an UAV with certain limitation and deficiency. If each sensor is independently utilized to estimate the attitude angles of an UAV, they can basically fulfill the purpose. But with time going by, some sensors might not respond fast enough, which may cause adverse data delay, while some sensors might be affected by the environmental factors and provide inconsistent measurements, yet some sensors might start providing inaccurate readings because sensor errors accumulate with time. All these behaviors will cause in-stability to the system performance and even lead to system failure. Therefore, a data fusion system which can adopt all the sensor readings, compensate the flaws of every sensor and possibly make the nearoptimal combination for attitude measurement is necessary to improve the flight performance of miniature UAVs for non-military low-cost applications. This paper reports our synergistic efforts in this regard. Attitude estimation is essential for UAVs to achieve autonomous navigation. In particular, for missions using UAVs carrying remote sensing payloads for geo-referencing purposes, accurate attitude estimation is more critical. For miniature UAVs, due to their weight and size limitations, it is easier for them to get affected by external disturbance, such as wind. Therefore, precise orientation data plays a more crucial role for the controller to stabilize the whole system and make the autonomous flight more safely and smoothly. In order to keep the overall system cost as low as possible while pursuing stable and accurate flight missions, combining all the available lowcost sensors by using a smart data fusion system is a preferred solution. This paper presents a low-cost data fusion system based on infrared (IR) sensor, inertial sensors and vision sensor. By integrating all three sensors’ data and feeding it into the proposed weighting filter based fusion system, it will analyze and avoid the deficiency of each sensor, make a near-optimal attitude estimation based on all the data comparisons. The major contribution of this paper is to provide a practical low-cost solution for accurate attitude estimation of miniature UAVs using different inexpensive sensors.
This paper is organized as follows: Section 2 introduces the basics of UAV attitude angles and compares three low-cost attitude estimation sensors. Then Section 3 explains the algorithms of these sensors for attitude estimations in our implementations. In Section 4, the weighting filter based sensor fusion system is presented. Section 5 shows the whole system implementation and preliminary test results. Section 6 concludes this paper and states the future plan.
2 Basics of UAV Attitude Estimation In order to achieve desired navigation performance, attitude estimation is the foundation. Attitude estimation with high fidelity will significantly improve the system stability and robustness. The most important parameters in UAV system states are the attitude angles, and they are also called the Euler angles when they are considered as angular rotations regarding the body fixed frame (Fig. 1). (1) Roll (φ): Rotation angle around the X axis of its body frame. (2) Pitch (θ ): Rotation angle around the Y axis of its body frame. (3) Yaw (ψ): Rotation angle around the Z axis of its body frame.
Fig. 1 Body frame and attitude angles
J Intell Robot Syst
The body frame is fixed to the UAV airframe, and the Euler angles rotate regarding to certain axes in this coordinate system so that the attitudes of the UAV can be measured. In this paper, restricted by the airframe design, the UAV system does not have all the conventional control surfaces like aileron, rudder and elevators. It has only two elevons, a combination of the aileron and elevator, so there is no dedicated control on yaw angles. Therefore, only the roll and pitch angles will be considered. The details regarding the airframe will be introduced later. There are several types of sensors in the market that have been used for UAV navigation purposes, such as inertial sensors, thermal sensors and vision sensors. Although they are based on different working principles and they have their own advantages and disadvantages, they can all provide attitude measurements and assist UAVs for autonomous navigation. Thermal sensors, which have been studied in [3], possess simple configurations with small cost. The most common thermal sensor is the IR sensor. It measures the infrared radiation emitted from objects having different temperature and compares the difference from its sensing surfaces. Based on the changes on its internal photosensitive material, the current and voltage of the sensor circuit also get changed, and then through a set of algorithms calculating those measurements, the UAV attitude angles can be estimated. Since the IR sensors generate analog signals, the sensor data can be fast sampled to achieve rapid estimation updates. However, because IR sensors heavily rely on the temperature factor which can change randomly in an outdoor area, its accuracy is hard to rely on if sometimes the weather changes apparently. Therefore, it usually involves some type of calibrations to rectify the sensors, and once the IR sensor is calibrated, its performance is still trustworthy. Inertial sensors include gyros and accelerators. Generally they are coupled to work together and become an inertial measurement unit (IMU). IMUs can provide fairly accurate attitude measurements because of their special configuration and complicated filtering algorithms. However, most IMUs available in the market are expensive and they are more popular on military UAVs,
commercial aircrafts and space shuttles. Therefore, they are not really applicable to civilian lowcost orientated projects. With the development of cheap inertial sensors in recent years, using some less sophisticated algorithm which requires smaller computational power, several low-cost IMUs containing three axes gyros, accelerometers and even magnetometers have become available [4]. Those IMUs work similarly to the commercial ones, but limit their price below one to two hundred US dollars [5]. Although their accuracy is not comparable to the industrial and commercial IMUs, they are suitable for the use of navigation missions in hobbyist and low-cost UAV projects. For IMUs, especially the low-cost ones, their gyros start having big drifts with time increase because the gyro bias is integrated over time due to cheap hardware. The attitude estimation will become inaccurate if the errors are accumulated to some extent and there are no other ways to rectify the sensor data. Once the IMU function fails, the UAV system will be jeopardized and even get crashed. Vision sensors, on the other hand for UAV navigations have become a popular topic recently. Video cameras are one of the most common vision sensors although there are other dedicated optical sensors, such as the Centeye visual microsensor [6]. Video cameras have been continuously improved in quality while being also more available for lower cost (less than $100 for high resolution). Based on advanced algorithms, video cameras can provide fairly accurate attitude estimations. Therefore, it can be used for UAV autonomous navigation or to verify the readings from other sensors. Another advantage of a vision-based navigation system is that it does not need GPS [7], which enables the system to be operated indoor or in areas which have problems receiving GPS signals. The biggest drawback of vision sensors for attitude estimation is that it needs plenty of computational power [8], and for a low-cost UAV project, it means another expense to invest on microcomputers or other hardware. In order to keep the overall cost as low as possible, the computational power has to be sacrificed, which means the sensor update rates cannot exceed a
J Intell Robot Syst Table 1 Sensor general comparisons
3 Sensor Attitude Estimation Algorithms
IMU
IR sensor
Video camera
Low cost Fast sensor updates Most accurate Sensor drifts
Low cost Fast sensor updates Least accurate Rely on temperature
Low cost Slow sensor updates Medium accuracy Require computation power
certain limit. However, since the video camera is able to provide reliable attitude estimation, it can be combined with IR sensors and IMUs to rectify their readings and improve the navigation performance of overall system. After the brief introductions of the three primary low-cost sensors, it can be concluded that IR sensor and IMU are both able to perform reasonable attitude estimation, and with GPS, they can individually serve as navigation unit for UAV autonomous flight. The video camera is also able to provide attitude measurement with medium fidelity. However, limited by the cost restriction, its sensor update rates cannot really support UAV navigation. IR sensors and IMUs also have obvious drawbacks, such as sensor drifting and influence from environmental factors. However, based on the smart data fusion system proposed in this paper, all the adverse features of every sensor can be possibly avoided and achieve the most accurate measurement from their combinations. A comparison of all the sensors introduced above is summarized in Table 1. In order to clarify the definition of low cost here, Table 2 shows a classification of IMUs in terms of price [5]. The lowcost standard defined in this paper is attributed to the Hobbyist Grade and it is actually less than $200.
3.1 IR Sensor Using IR sensors for UAV attitude estimation has been achieved in [9], in which IR sensors and GPS are utilized for autonomous navigation and detailed analysis of IR sensors’ characteristics regarding orientation angles are provided. Usually IR sensors need to work in a pair, so one behaves as the vertical channel to cover the z-axis and the other one behaves as the horizontal channel to cover the x-axis and y-axis. Figure 2 shows an example of how to install two IR sensors on a fixed wing UAV. Most IR sensors consist of two to four sensing ports, and as it is illustrated above, the vertical sensor only uses two sensing ports while the horizontal sensor uses all four sensing ports. Each of the sensing ports will receive infrared radiation and compare with each other, then derive the attitude angles. But in this paper, we will focus on
(a)
Table 2 IMU categories IMU type
Cost ($)
Example
Navigation grade Tactical grade Industrial grade Hobbyist grade
>50 k 10–20 k 0.5–3 k