Скачать с ютуб One-Shot Learning from Demonstration в хорошем качестве

One-Shot Learning from Demonstration

This project presents a novel approach for robot arm path planning called One-Shot Learning from Demonstration (LfD), applied explicitly to tree branch-cutting tasks. The method utilises a one-shot recognition technique to learn the cutting route from a single demonstration by an expert (farmer). In this method, a pose detection algorithm first tracks the demonstrator’s index fingertip to capture the cutting trajectory, while YOLO object detection extracts key features from the demonstration video. These features include the average and standard deviation of the positions of leaves and fruits, representing their spatial distribution along the branch. A neural network then maps these features and the cutting path into an embedded space, facilitating one-shot learning from the human demonstration. Once trained, the system can generate an optimal cutting route for any configuration of leaves and fruits. The approach was tested in both simulated and real-world environments, demonstrating excellent performance in reproducing the cutting route based on a single demonstration. #RobotPathPlanning #OneShotLearning #LfD #RobotArm #TreeCutting #PoseDetection #YOLODetection #AIInAgriculture #RobotLearning #BranchCutting #NeuralNetworks #OneShotRecognition #PrecisionFarming #Agritech #AIInnovation