High-Accuracy Position-Aware Robot for Agricultural Automation Using Low-Cost IMU-Coupled Triple-Laser-Guided (TLG) System

A novel, low-cost approach to high-accuracy robot localization for agricultural applications using an image-processing triple-laser-guided (TLG) system coupled with an inertial measurement unit (IMU) is presented in this paper. The TLG system consists of a laser-pointing unit (LPU) at the base station and a laser-target unit (LTU) at the mobile robot. The robot’s XYZ position and heading are determined from the positions and the angles relative to the field of both LPU and LTU. The robot’s roll and pitch are determined by the IMU sensor fusion with complementary filter. The IMU-coupled TLG system is demonstrated on an outdoor, $20 \times 21$ m flat field at various light intensities. The overall lateral and vertical accuracies of the IMU-coupled TLG system are 1.68 cm and 0.59 cm, respectively. The overall heading, roll, and pitch accuracies of the IMU-coupled TLG system are 0.90°, 0.78°, and 0.76°, respectively. The lateral and heading accuracies of the IMU-coupled TLG system are found to be comparable to commercially available GNSS-INS systems from NovAtel and Trimble, while the total cost of the IMU-coupled TLG system is only a fraction of the total cost of the commercially available localization systems.


I. INTRODUCTION
Automatic machinery has been widely researched and applied in the agricultural industry to increase productivity and reduce costs. Specifically, automatic guided robots are of interest for many agricultural tasks, i.e., seeding, weed controlling, soil tilling, fertilizing, watering, and harvesting. To effectively perform these complex open-field tasks, accurate and precise localization of these agricultural robots, including XYZ position, heading, and attitude (roll and pitch tilts), are essential. Several principles of localization sensors for agricultural robots are real-time kinematic global navigation satellite systems (RTK GNSS) [1]- [5], landmarks detection [6], light laser detection and ranging (LiDAR) [7]- [8], radar [9], sensor fusion with camera [10]- [13], ultrasonic [14], multi-sensor fusion [15], [16], beacons based [17], localization and mapping (SLAM) [18], laser range finder (LRF) [19], and inertial measurement unit (IMU) [20]. Particularly, RTK GNSS coupled with inertial navigation The associate editor coordinating the review of this manuscript and approving it for publication was Yang Tang . systems (INS) is extensively employed by farmers due to its relatively higher accuracy and precision compared to other sensing techniques. In the RTK GNSS-INS system, the robot's position is determined based on the trilateration of the Universal Transverse Mercator obtained through a constellation of satellites, while the robot's angles of orientation are obtained by real-time measurements of angular velocity, linear acceleration, and the earth's magnetic field via the INS.
Nonetheless, there are a few drawbacks to the agricultural practicality of the RTK GNSS-INS system. First, the RTK GNSS-INS system's accuracy and precision depend on the transmissivity of satellites' signals. The decrease in the satellite transmissivity due to signal blockage and multipath effect from neighboring buildings or trees and cloudy weather can lead to the deterioration of the RTK GNSS-INS system's positioning accuracy and precision. Hence, this limits the operation of RTK GNSS-INS system to only an open, outdoor environment under open-sky conditions. Second, the heading determination in the RTK GNSS-INS system partly relies on the sensor fusion between the earth's magnetic field measurement from magnetometers and the angular velocity measurement from gyroscopes with tilt compensation accelerometers in the INS unit. Due to geographical variation in the earth's magnetic field, the RTK GNSS-INS system requires a time-consuming calibration when the operating field has been moved. Moreover, the measurement of the earth's magnetic field can easily be interfered with by the surrounding environment, such as metallic structures and electronic equipment; therefore, the INS's heading measurement can, at times, be inconsistent and unreliable.
In recent years, there are advancements in GNSS-based localization systems for agriculture to tackle the previously mentioned drawbacks. For example, integration of RTK GNSS-based system with LiDAR [21] and Untethered Dead Reckoning (UDR) [22] have been shown to improve the accuracy and precision while operating in GNSS-denied areas. In some applications, multi-antenna GNSS systems were shown to improve the heading accuracy and RTK availability/success rate in signal-challenging areas [23]- [25]. However, all of these could lead to increase cost and complexity to the end users significantly.
Lastly, the RTK GNSS-INS system is currently costly and not affordable to small-scale farming operations. To provide an alternative to the conventional, expensive RTK GNSS-INS system, this paper presents a low-cost, inertial measurement unit (IMU)-coupled triple-laser-guided (TLG) system that provides high localization accuracy and precision for both indoor and outdoor environment at a substantially lower cost.

II. BRIEF INNOVATION DESCRIPTION
The IMU-coupled TLG system consists of a base station with a laser-pointing unit (LPU) and a mobile unit with a lasertarget unit (LTU). The LPU consists of two laser pointers and a laser rangefinder module. The LTU consists of three laser targets: two projection targets to detect the projected laser pointers and one target to reflect the laser rangefinder from the LPU. The high accuracy and precision of localization are achieved by the innovative design of interaction between the LPU and LTU which is briefly described as follows: -The LPU is controlled to point at the middle of the LTU while the LTU is controlled so that the laser targets are perpendicular to the laser-pointing beams. -The interactive controlling of the LPU and LTU is achieved by analyzing the projected positions of the LPU's laser pointers on the LTU's projection targets via image processing. -With the LPU pointing perpendicularly at the center of the LTU, the position and heading of the mobile unit relative to the laser base station can be accurately and precisely determined using the distance measured by the laser rangefinder, the fanning angle of LPU, and the relative heading angle of the LTU.

III. EQUIPMENT DESCRIPTION
The IMU-coupled TLG system demonstrated in this study consists of a laser base station and a mobile unit, as shown in Fig. 1. Computer software is used for data recording, manual control, and PID setup for both the base station and the mobile unit.

A. LASER BASE STATION
The laser base station comprises a base frame, a laserpointing unit (LPU), and a fanning-vertical control unit (FVCU), as shown in Fig. 2. The base frame is constructed as a rectangular box frame with dimensions of 30 cm in width, 50 cm in length, and 110 cm in height, using hollow steel bars. Four level-adjusting poles are applied as the base frame's legs.
The LPU is designed to track the position and control the heading of the mobile unit. The LPU holds three lasers: a heading-measuring laser (HML), a position-measuring laser (PML), and a distance-measuring laser (DML). The HML, the PML, and the DML are at the top, in the middle, and at the bottom position of the LPU, respectively, with a 12-cm vertical separation. The lasers are situated to point in the same direction and parallel to the ground. A 405-nm violet laser pointer, a 650-nm red laser pointer, and a 650-nm red laser rangefinder are applied as the HML, the PML, and the DML, respectively.
The FVCU, devised to track and control the linear and fanning motions of the LPU, consists of a rotary actuator and a linear actuator (Fig. 2). The rotary actuator employs a micro-stepping motor and an incremental encoder with a maximum angle of rotation of 180 • and a rotating resolution of 9.38 × 10.3 • /step. The linear actuator is driven by a DC motor and controls its moving arm position using an incremental encoder and PID position control with the maximum moving distance of 30 cm. In the FVCU, the rotatory actuator is attached to the moving arm of the linear actuator. It is oriented so that its plane of rotation is perpendicular to the linear actuator's uniaxial motion.
To allow vertical control along the Z-axis and fanning control in the XY-plane of the LPU, the LPU is attached to the rotary actuator of the FVCU, and then the LPU-FVCU assembly is secured on the base frame so that the linear actuator of the FVCU is in the vertical position. A microcontroller unit (MCU) is employed to process the data from the FVCU and the LPU, to control the actuators in the FVCU, and to remotely communicate the data via 433-MHz radio frequency (RF).

B. MOBILE UNIT
The mobile unit consists of a skid-steering robot, a laser-target unit (LTU), a heading control unit (HCU), and an attitude-measuring unit (AMU), as illustrated in Fig. 3. The skid-steering robot is a four-wheel-drive using four 24-volt DC motors with H-bridge PWM drivers controlled by an MCU. The skid-steering robot's commands from an external host are remotely transmitted to the MCU via 433-MHz RF. The robot's frame dimensions are 50 cm in width, 70 cm in length, and 140 cm in height, and the size  of the wheels is 20.4 cm in diameter. The robot's maximum speed is set at 1.5 m/s.
The LTU includes a heading-measuring target (HMT), a position-measuring target (PMT), a distance-measuring target (DMT), and an image sensor. The HMT and the PMT, assigned to project the HML and the PML for heading control and measurement, are 12 cm × 50 cm projector screens made from black, translucent paper. The DMT, assigned to reflect the DML for distance measurement, is a 12 cm × 50 cm laser reflecting screen made from a white, opaque plastic sheet. The HMT, the PMT, and the DMT are placed at the top, in the middle, and at the bottom position of the LTU, respectively, with a 12-cm vertical separation between their centers. The image sensor used to detect and determine the positions of the projected HML and PML on the HMT's and PMT's screens is a Pixy2 CMUcam5 smart camera with a frame rate of 60 frames per second and the field of view (FOV) covering both HMT and PMT. The HMT, the PMT, and the image sensor are placed in a dark enclosure, where the HMT and the PMT are placed in front of the image sensor at 18 cm and 30 cm away, respectively.
The positions of the projected HML on the HMT and the projected PML on the PMT are used to control the LPU and LTU so that the lasers perpendicularly pointing at the center of their assigned targets. This is critical for the IMU-coupled TLG system to achieve the highest accuracy and precision in localization and is elaborated in further detail in Control Description. The Pixy2 smart camera is programmed to find a violet laser blob from the HML on the HMT and a red laser blob from the PML on the PMT. The centroids of the area of these laser blobs are then quantified via image processing of the Pixy2 smart camera and reported to the microcontroller as the positions of the projected lasers. To avoid interference for the image sensor's detection of projected HML and PML, the maximum light intensity allowed on the HMT and the PMT is 20,000 lux.
The HCU is a 360 • bidirectional, rotatable stand. A DC-motor rotary actuator with PID position control drives and controls its rotational motion and an absolute encoder measures its angle of rotation. For attitude (roll and pitch tilts) measurement, the AMU utilizes a low-cost 6DoF IMU where roll and pitch are computed using IMU sensor fusion with complementary filter [26], [27].
For the complete assembly of the mobile unit, the HCU is secured on top and at the center of the robot's frame. The LTU and AMU are then placed and centered on the HCU, allowing 360 • rotation of the LTU and the AMU on the mobile unit. An MCU is utilized to process the data from the LTU, the HCU, and the AMU, control the rotary actuator in the HCU, and remotely communicate the data via 433-MHz RF.

IV. SYSTEM COST BREAKDOWN
The IMU-coupled TLG system's total cost mainly comes from the structural, mechanical, electronic components of three hardware groups: the laser base station, the mobile unit, and the wireless communication components. The cost breakdown was based on retail and online prices plus shipping for Bangkok, Thailand, in February 2021 and is summarized in Table 1. The total cost of the IMU-coupled TLG system for this study was $950 in USD. It is to be noted that the skid-steering robot's cost is not included since its purpose in this study was only to be a model rover for demonstrating the performance of the IMU-coupled TLG system.

V. WORKING FIELD AND REFERENCE AXES
An operational area for the mobile unit is a 20 m × 21 m rectangular plain with its reference axes, designated as X, Y, and Z, defined with respect to the laser base station, as depicted in Fig. 4. The origin (0,0,0) of the working field is assigned as the LPU's position at the laser base station. Due to the DML's low measurement accuracy at a close distance to the target, the working field excludes a semicircular area with a radius of 1 m around the laser base station.

VI. INPUT PARAMETERS
The input parameters for determining the robot's position and orientation (attitude and heading) in the IMU-couple TLG system, assuming that both the LPU and LTU are correctly and promptly controlled as elaborated in Control Descriptions, are described as follows: 54328 VOLUME 9, 2021 1) The distance between the LPU and the LTU (r) measured by the DML on the LPU at the laser base station (see Fig. 4) 2) The fanning angle of the LPU (β) measured by the FVCU at the laser base station (see Fig. 4) 3) The height of the LPU (h) measured by the FVCU at the laser base station 4) The roll tilt of the LTU (θ) measured by the AMU on the mobile unit 5) The pitch tilt of the LTU (φ) measured by the AMU on the mobile unit 6) The relative heading angle between the LTU and the skid-steering robot (γ ) measured by the HCU on the mobile unit 7) The robot's heading on the field (α) is the sum of β and γ

VII. CONTROL DESCRIPTION
In the IMU-coupled TLG system, the robot's position and heading are computed using the distance between the LPU and LTU and the height and fanning angle of the LPU, which is further elaborated in Determination of Robot's Position and Orientation. For the highest accuracy of distance and heading measurements, the lasers, namely the HML, the PML, and the DML, must be parallel to the ground and perpendicularly pointing at the center of their assigned targets, the HMT, the PMT, and the DMT, respectively. Hence, the LPU on the laser base station and the LTU on the mobile unit must be simultaneously controlled as follows:

A. FANNING ANGLE AND HEIGHT CONTROL OF THE LPU
The LPU control, carried out by the FVCU, aims to keep its lasers' projection at the center of their assigned targets. Since the HML, the PML, and the DML are all on the same LPU's rigid body and positioned with the same vertical separation as the centers of their assigned targets on the LTU, the LPU control is solely focused on maintaining the projected PML position at the center of the PMT. If the PML points perpendicularly to but away from the center of the PMT, it yields center-deviated distances of the projected PML in both X-and Z-axes of the PMT, designated as Err PMT X and Err PMT Z , respectively, as illustrated in Fig. 5(a). To keep the projected PML at the PMT's center, the rotary actuator in the FVCU controls the fanning angle (β) of the LPU to minimize the Err PMT X to zero, while the linear actuator in the FVCU controls the height (h) of the LPU to minimize the Err PMT Z to zero, as depicted in Fig. 5(b). The LPU fanning angle and height control for Err PMT X and Err PMT Z minimization is achieved using a PID control with PWM signaling to the FVCU, as illustrated in Fig. 5(c).

B. HEADING CONTROL OF THE LTU
The aim of the LTU's heading control carried out by the HCU is to ensure that three lasers' beams from the LPU are perpendicular to the targets of the LTU. Suppose the LPU's laser beams are non-perpendicular to the LTU's targets' screens. In that case, there is a difference in the center-deviated distances along the X-axis of the targets between the projected HML on the HMT (Err HMT X ) and the projected PML on the PMT (Err PMT X ), as shown in Fig. 6(a). The magnitude of the difference between Err HMT X and Err PMT X , designated as Err X , is defined as follows: If Err X > 0, the rotary actuator in HCU rotates the heading of LTU to minimize Err X to zero, confirming the laser beams' perpendicularity from the LPU to the LTU's targets, as illustrated in Fig. 6(b). The LTU heading control for Err X minimization is achieved using a PID control with PWM signaling to the HCU, as illustrated in Fig. 6(c).

VIII. DETERMINATION OF ROBOT'S POSITION AND ORIENTATION
Assuming the perpendicularity and centering of the lasers' projections on the targets as described in Control Description, the robot's position and heading in Fig. 7 can be accurately determined as follows: Heading(α) = β + γ The robot's roll (θ) and pitch (φ) tilts are calculated via sensor fusion of low-frequency signals from a triple-axis VOLUME 9, 2021  accelerometer and high-frequency signals from a triple-axis gyroscope of a low-cost 6DoF IMU in the AMU through complementary filter [26], [27].

IX. DATA FLOW AND COMMUNICATION DIAGRAM
In the IMU-coupled TLG system, the laser base station, the mobile unit, and the computer communicate data among themselves with a sampling rate of 100 ms, as depicted in Fig. 8. The communication is carried out using HC-11 wireless transceiver modules with a frequency band of 433 MHz. For all transceiver modules, the RF power is set to 10 dBm with the data baud rate of 115,200 bps and the transmission delay time of 10 ms. The operating distance of these communication modules is approximately 30 m.

X. METHOD FOR SYSTEM DEMONSTRATION
The operation of a high-accuracy, position-aware robot using the IMU-coupled TLG system was demonstrated in an outdoor setting on a 20 m × 21 m treeless, flat field, as depicted in Fig. 4. The robot was manually controlled to move along a predetermined path on the field from the numbered location 1 to 10, as shown in Fig. 9, without exceeding the maximum speed of 1.2 m/s. The robot's heading was preassigned at the numbered locations with a designated arrow. At each numbered location, the robot was stopped to physically verify its actual position (XYZ coordinates) and angles of orientation on the field, and the reported values from the IMU-coupled TLG system were recorded. The light intensity on the laser targets, HMT and PMT, was also noted at each numbered location. Furthermore, the robot's XY position was recorded every one second by both systems to generate its horizontal trajectory on the designed path. To validate the system's outdoor capability in daylight and ensure the stringency of the  statistical analysis, the system demonstration was repeated five to ten times at each numbered location under various light intensities on the field ranging from 45,000 lux to 85,000 lux during the day time.
The robot's actual position and orientation (R l ) at the l th location is defined as: The system's reported (R li ) position and orientation of i th epoch at the l th location is defined as: VOLUME 9, 2021 TABLE 2. The actual position and orientation of the skid-steering robot physically verified at each numbered location on the field.
The system's average reported position and orientation (R l ) and the standard deviation (σ l ) at the l th location are computed as follows:R (8) where N l is the total number of epochs at the l th location.
The system's position and orientation accuracy (A l ) at each numbered location are reported as the absolute difference between the robot's actual position and orientation (R l ) and the system's average reported values (R l ), which can be expressed as: By averaging A l from all numbered location, the overall system's position and orientation accuracy (A OA ) can be determined as follows: where M is the total number of the measured locations. The system's position and orientation precision at each numbered location is reported as the standard deviation (σ l ) of the measurements at that numbered location. The system's overall position and orientation precision (P OA ) are quantified as the combined standard deviation of the system's reported position and orientation at all numbered locations. Since the acquired measurement at each location is independent of one another, the standard deviations were combined using a weighted sum of variance. Hence, the system's P OA is expressed as:

XI. RESULT AND DISCUSSION
For the determination of the system's accuracy, the actual position and orientation of the robot and the light intensity on the HMT and the PMT at each numbered location were physically verified and reported, as summarized in Table 2. After multiple measurements at each numbered location, the robot's average XYZ coordinates, headings, and attitudes with the computed local accuracy reported by the IMUcoupled TLG system are shown in Table 3. The overall accuracies for X, Y, Z, α, θ, and φ reported by the IMU-coupled TLG system are within 0.92 cm, 1.37 cm, 0.59 cm, 0.90 • , 0.78 • , and 0.76 • , respectively. The local maximum errors and standard deviations at each numbered location reported by the IMU-coupled TLG system are presented in Table 4. The overall precisions for X, Y, Z, α, θ, and φ reported by the IMU-coupled TLG system are within 0.75 cm, 0.69 cm, 0.67 cm, 0.96 • , 0.71 • , and 0.59 • , respectively. The overall maximum errors for X, Y, Z, α, θ, and φ reported by the IMU-coupled TLG system are 3.00 cm, 4.00 cm, 2.00 cm, 3.00 • , 3.00 • , and 2.20 • , respectively. Overall maximum error, accuracy, and precision of the robot's position and orientation reported by the IMU-coupled TLG system are summarized in Table 5. Moreover, a multipass, horizontal trajectory of the robot generated by real-time XY position tracking via the IMU-coupled TLG system is depicted in Fig. 10. Hence, the localization and the horizontal path tracking of the IMU-coupled TLG system were found to be highly accurate and precise all along the designed path. Since the XYZ and heading measurements by the IMU-coupled TLG system solely rely on the accuracy of HML and PML detection on the HMT and PMT, the laser beam divergence due to its travel distance and sunlight interference could deteriorate the overall accuracy and precision. The local XYZ/heading accuracy and precision at each numbered location were plotted against the mobile unit's distance from the laser base station and the average light intensity on the laser targets. The local XYZ/heading accuracy and precision versus the robot's distance from the laser base at each numbered location are illustrated in Fig. 11(a)-11(d) and Fig. 11(e)-11(h), respectively. Within the given field size, no correlation between the IMU-coupled TLG system's overall accuracy and precision and the mobile unit's VOLUME 9, 2021 TABLE 5. Summary of overall maximum error, accuracy, and precision of the robot's position and orientation reported by the IMU-coupled TLG system.  distance from the laser base station was established. The local XYZ/heading accuracy and precision versus the average light intensity on the laser targets at each numbered location are depicted in Fig. 12(a)-11(d) and Fig. 12(e)-11(h), respectively. If not exceeding 20,000 lux, the varying light intensity on the laser targets during the daytime did not significantly affect the system's overall accuracy and precision.
The overall localization performance and the total cost of the IMU-coupled TLG system are compared to those of the commercially available GNSS-INS systems, namely NovAtel PwrPak7D-E2 and Trimble BX992. The performance comparison criteria are lateral accuracy (A LAT OA ) and heading accuracy (A α OA ). The lateral accuracy of the IMU-coupled . The local accuracy (A l ) and precision (σ l ) of the robot's XYZ coordinates and heading reported by the IMU-coupled TLG system at the numbered locations versus the mobile unit's distance from the laser base station.
TLG system is computed from the overall XYZ accuracies (A X OA and A Y OA as shown in Table 5) using the equations as follows: The comparison of the IMU-coupled TLG system with the NovAtel PwrPak7D-E2 and Trimble BX992 is summarized in Table 6. The lateral and heading accuracies of NovAtel PwrPak7D-E2 and Trimble BX992 are based on the best performance claimed by their system specifications. The lateral accuracies of the IMU-coupled TLG system, NovAtel PwrPak7D-E2, and Trimble BX992 are 1.68 cm, 1.00 cm, 5.00 cm, respectively. The heading accuracies of the IMU-coupled TLG system, NovAtel PwrPak7D-E2, and Trimble BX992 are 0.90 • , 0.05 • , and 0.09 • , respectively. This shows that the localization performance of the IMU-coupled TLG system is comparable to those of the commercially available GNSS-INS systems. Furthermore, while the GNSS-INS systems are limited only for an open, outdoor environment under optimal open-sky conditions, the IMU-coupled TLG system can operate in both indoor and outdoor settings with the only outdoor limitations of rainy weather and maximum light intensity of 20,000 lux.
The total cost of the IMU-coupled TLG system is approximately $950 in USD, while those of NovAtel PwrPak7D-E2 and Trimble BX992 are approximately $11,000 in USD and $6,000 in USD, respectively. The costs of the NovAtel PwrPak7D-E2 and Trimble BX992 are based on the actual online prices in February 2021, which may vary depending on the location. Also, there is an additional yearly cost for GNSS correction services, i.e., for this instance, TerraStar for NovAtel GNSS receivers and Trimble-RTX for Trimble receivers. The prices and availability of the GNSS correction services may vary depending on the geographical location. If the GNSS correction service is unavailable in the desired area of operation, a separate GNSS base is required, which may add up to $12,000 in USD into the total cost of the GNSS-INS system. This can be concluded that the total cost of the IMU-coupled TLG system is at least five times less expensive compared to the commercially available GNSS-INS systems in the current market.

XII. CONCLUSION
After repeated robot's position and orientation measurements on a 20 × 21 m flat, open field, the IMU-coupled TLG system with a total cost of only $950 in USD reported the lateral and heading accuracy of 1.68 cm and 0.90 • , respectively. The lateral and heading accuracy performances are comparable to those of commercially available GNSS-INS localization systems. The multipass horizontal path tracking also confirmed the high accuracy and precision of the IMU-coupled TLG system. Furthermore, the IMU-coupled TLG system reliably performed with various light intensities on the laser targets under 20,000 lux without any noticeable deterioration in localization accuracy and precision. Hence, this novel IMU-coupled TLG system presents a new promising solution as a low-cost, high-accuracy alternative for automatic machinery in the agricultural industry.