Development of a Robotic Pruning System for Sustainable Apple Production

Objective #1:  Development of an End-effector for Robotic Pruning

1. Dynamic Analysis for Branch Pruning

The amount of force required for cutting branches is an important parameter for the pruning end-effector design. A branch pruning dynamic analysis was performed to measure the cutting torque requirements for different apple cultivars (Golden Delicious, Gala, Fuji, and Honeycrisp). The effect on cutting torque requirements was also investigated on branch cutting angle and branch-blade contact point. To measure the branch cutting torque, a force measuring sensor was integrated with a manual shear pruner (Figure 1). The cutting torque was calculated using the measured force and the length of the pruner arm. In the test, 630 cuts were made with different diameters. The maximum force for each cut was recorded, as well as the corresponding branch diameter. An Inertial Measurement Unit (IMU) was also used to monitor the angle of shear blades with the branch.

Figure 1. Manual pruner equipped with a force sensor for pruning force measurement

2. Pneumatic End-effector Development

A concept design of a robotic pruning end-effector was developed in SolidWorks environment (Figure 2). Considering the needed maneuverability of the end-effector, the concept design includes two rotational joints and a cutter which consists of a cutting blade and an anvil. The cutting mechanism was operated using the pneumatic system. The simulation was also conducted in Matlab software for workspace analysis of the end-effector, as well as the capability of aligning with the branches at different orientations.

Figure 2. A two DoF robotic pruning end-effector. 1, 2- Motor, 3, 6- Gearbox, 4- Cutter, 5- Pneumatic cylinder

3. Experiment of the Integrated Pneumatic End-effector

The field test for the validation of simulation results and performance assessment of the end-effector was conducted on tall spindle ‘Fuji’ apple trees at Penn State Fruit Research and Extension Center (FREC, Biglerville, Pennsylvania). Five ‘Fuji’ trees were selected randomly, and 15 to 20 branches of different orientations were selected from each tree for tests (Figure 3). The pressure of the pneumatic cylinder was set at 0.8-0.825 MPa. In test one, the pneumatic cylinder with a bore size of 2.2 cm was used. In total, 100 cuts were tested as multiple cuts were applied to each selected branch at different locations and diameters. In test two, the pneumatic cylinder with a bore size of 2.54 cm was used to provide a larger cutting force. A similar test was conducted on randomly selected trees and 120 cuts were tested including multiple cuts to each selected branch at different diameters and orientations on selected branches.

Figure 3. The developed robotic end-effector integrated with the three directional linear actuators

Objective #2: Development of An Integrated Manipulator System

1. Three Degrees of Freedom Pruning End-effector

The end-effector was designed with an intrinsic three-rotational (3R) degrees of freedom (DoF) configuration (Figure 4). A computer-aided design (CAD) software, named SolidWorks, was used as a design tool. One of the primary objectives of the work was to minimize the spatial requirements of the robot during maneuvering in the tree canopy. The design criteria including manipulability, mechanical strength, and spatial requirements were satisfied using the Motion Simulation Toolbox in SolidWorks. An electric shear blade cutter was developed and attached to the last joint of the end-effector. To avoid joint collisions, the position/placement of the joints and shear cutter were selected based on the motion simulation results.

Figure 4. Concept design of the end-effector. (components: 1. Motor for yaw rotation; 2. Motor for pitch rotation; 3. Motor for roll rotation; 4. Self-locking worm gearbox; 5. Shear cutter)

2. Integrated Pruning Manipulator

A three-prismatic (3P) DoF cartesian manipulator was designed in SolidWorks for the integration of the 3P DoF end-effector. A rigid linear arm was designed to integrate both systems to develop an integrated 6 DoF robotic pruning system (Figure 5). SolidWorks motion simulation toolbox was used to validate the design for joint collision and assess the performance of the system.

Figure 5. Integrated pruning system with an end-effector attached to a cartesian manipulator; the components include: 1) x-axis rails; 2) y-axis rail; 3) z-axis linear actuator; 4) axis limit switches; 5) linear arm; 6) pruning end-effector

3. Experiment of the Integrated Pruning System

The field tests were conducted on trellised ‘Fuji’/ Bud. 9 apple trees trained to a fruiting wall architecture at Penn State’s Fruit Research and Extension Center, Biglerville, Pennsylvania (Figure 6). About 100 cuts were applied on branches at a wide array of orientation ranges. A set of ten apple trees were selected randomly from the same orchard block. For each tree, about ten branches of different positions and orientations were selected. The coordinates of the cut points and the orientation of branches were estimated manually from the fixed manipulator base and entered through the GUI. Some cuts were applied approximately 10 to 20 mm from the tree trunk to evaluate the end-effector cutter capability to prune the branches close to the trunk. The branch diameter at the cut point and corresponding joint positions were recorded to validate the design parameters of the integrated pruning system.

Figure 6. Experimental setup of the integrated pruning system in the apple orchard

Objective #3: Investigation of Branch Accessibility with a Robotic Pruner

The spatial requirements of a robotic arm could be estimated based on the location of the branch to establish the trajectory of the robotic arm. Canopy characteristics such as branch density and branch dimension affect the path of the robotic arm to reach the object. The study includes the following tasks: 1) Reconstruction of apple tree; 2) Establishing simulation environments in Matlab; 3) Developing and implementation of collision-free path planning algorithms.

1. Reconstruction of Apple Tree

A Three-dimensional (3D) model of the Fuji’ apple tree having tall spindle tree architecture was developed using the point cloud data acquired from the VLP-16 LiDAR scanner (Velodyne LiDAR, San Jose, CA, USA). All secondary branches from the tree were removed before data collection as these were not considered rigid obstacles. The point cloud data were processed using the function velodyneFileReader() in Matlab (MathWorks Inc, MA, USA). The developed tree model consists of a trunk and 14 primary branches, which were reconstructed using the function spline() in Matlab (Figure 7). The diameter of the trunk and branches were added manually to reconstruct the final tree structure.

Figure 7. Reconstruction of the apple tree using LiDAR point cloud data

2. Establishing Simulation Environments

A 6R DoF robotic manipulator system: The robotic manipulator used was a 6R DoF industrial robotic arm (UR-5, Universal Robots, Odense, Denmark). The tree model and kinematic model of the manipulator were imported in Matlab to develop the simulation environment.  The base of the manipulator was set as origin i.e., x= 0 mm, y=0 mm, and z= 0 mm. Based on the allowable workspace of the manipulator and the physical parameters of the tree model, the distance between the manipulator base and the tree trunk was set at 400 mm (Figure 8(a)).

A 3R3P DoF robotic manipulator system: The kinematic model of the 3R3P DoF manipulator was developed by calculating the DH parameters of the robot. A simulation environment was established in Matlab, comprised of an apple tree model and kinematic model of the robot. The manipulator was set as origin at x= 0 mm, y=0 mm, and z= 0 mm, and the apple tree model was added to the simulation environment with trunk positioned at x = 1000 mm, and y = 250 mm (Figure 8(b)).

For both simulation environments, the task space cubic polynomial trajectory solution was established based on the inverse solution involving minimum joint displacement to move the manipulator from the start to the desired position.

Figure 8. (a) Cutter approach posture at the target branch

Figure 8. (a) Cutter approach posture at the target branch

3. Developing and Implementation of Collision-free Path Planning Algorithms:

Path Planning Algorithms: An obstacle avoidance algorithm using Rapidly-exploring Random Tree (RRT) was developed for a collision-free path to reach the target pruning points. The RRT algorithm performs two checks: manipulator body collision and end-effector path collision. Target branch coordinates were added and if the algorithm found no collision, the specific path nodes are added to the final solution, and the process continues until connected nodes reach the target location. Furthermore, RRT path smoothing and optimization algorithms were also used to reduce path length and calculate the optimized path. The RRT path smoothing aims to reduce the path length by removing unnecessary path nodes. For path optimization, a non-linear optimization algorithm was implemented with initial and boundary conditions. The minimum avoidance distance from the obstacles was set to 60 mm.

Path Implementation: The simulation setup was created on a laptop computer with Windows 10 operating system (Microsoft Inc., USA) and 16 GB memory. For both robotic systems, the simulation was performed to reach different target pruning points in the apple tree. The coordinates of target pruning points on each branch were marked 20 mm away from the tree trunk. The simulation was conducted by adding multiple approaching poses of the cutter at the target branch (Figure 9(). For each simulation, three branches were selected as target pruning branches. The data were recorded for path finding success, path length, and computational time for all path planning methods to compare the performance for different algorithms. In addition, a lab validation test for a physical 6R DoF robotic arm was also conducted. Serial communication was established between the simulation and the robot controller.

Figure 9. Cutter approach posture at the target branch