Project: Autonomous Apple Harvesting via Reinforcement Learning
Description
VBTI is an AI engineering company specializing in developing Deep Learning solutions for industries such as agriculture and manufacturing.
This project aims at developing Autonomous Apple Harvesting, building on an existing proof-of-concept previously created by the company. The initial implementation utilized an object detection model to identify apples and a depth camera to localize their positions. The harvesting process was simulated by directing the robot arm to the detected apple’s location. Traditionally, such tasks require motion planning to be explicitly programmed for each specific scenario. The objective of this project is to explore an alternative solution based on Reinforcement Learning by training motion control policies directly from visual inputs provided by the camera. This method eliminates the need for manual programming, enabling a more generalized and adaptable approach to robot motion control.
The recently launched Genesis (a physics platform tailored for general-purpose robotics, embodied AI, and physical AI applications) will serve as the foundation for this work. This innovative platform facilitates the training of motion control policies across a variety of robotic systems. Our initial focus will be to model the apple harvester within the simulation environment and train a reinforcement learning agent specifically designed for apple harvesting. The subsequent step will involve applying transfer learning techniques to adapt the trained model for deployment on a physical robot.
If successful, this approach could be applied to several other use cases.
Please contact Illya Kaynov (illya.kaynov@vbti.nl) for more information.
Details
- Supervisor
-
Maryam Tavakol
- Secondary supervisor
-
IKIllya Kaynov
- External location
- VBTI