Motion planning algorithm and its real-time implementation in apples harvesting robot

A fruit-by-fruit harvesting robot has to reach the target fruits on a tree and perform the detaching movement without any collision with the obstacles (trunks, branches, leaves, etc.) in its workspace. However, the working conditions in a fruit orchard are only partially structured and partially visible. In such cases, a flexible motion planning algorithm, which allows replanning of the ongoing executed motion while receiving additional information from the environment, is required. In this study, an anytime motion planning algorithm for a 9 degree-offreedom (DoF) manipulator to be used for selective apples harvesting and its real-time implementation are presented. Among the existing motion planning algorithms in literature, the sampling-based approaches are widely applied in robotics due to their capabilities of dealing with many DoFs robots without an explicit representation of the obstacles. In the proposed algorithm, the motion planning task is an ongoing process where the motion is re-planned in every configuration of the trajectory based on the acquired obstacle data. This approach requires a robust and time-efficient motion planner. Therefore, several different samplingbased motion planning algorithms have been evaluated in a specific simulation environment which is similar to the real working conditions to evaluate the most efficient planner. The proposed re-planning approach has been integrated by considering two parallel threads for planning and execution. The planning thread constructs an obstacle free roadmap and finds the trajectory for the user requested query. The execution thread takes the trajectory from the planning, sends the velocity references to the low-level controllers and monitors any change in the user request or new detected obstacles blocking the current executed trajectory. From this monitoring and the pre-defined delay of the trajectory execution, an updated query is generated and sent to the planning thread. This enhancement allows the planner to deal with not only any changing environment including newly observed obstacles or moving obstacles, but also any change in the high-level task which demands a new target location. The effectiveness of the proposed method has been validated in simulation where new obstacles appeared on the original trajectory or the target was changed to another location while executing the predefined path. The proposed method has also been implemented into a real manipulator by using an adaptive delaying buffer in motion command execution. Both in simulation and in the real-time tests the planner handled every situation correctly within an acceptable computation time.