Description

This project will be build upon an existing perception system that utilizes an RGB-D camera attached to a Kinova manipulator. A next-best-view (NBV) algorithm will be developed to optimize the approach to a cotton boll, taking into account the distribution of the cotton plant and confirming the class membership of the target instance. Additionally, the system will be capable of harvesting the cotton boll properly once its class is confirmed.

The project will involve the creation of a detection and segmentation model specifically tailored for the objects of interest. Furthermore, it will include the development or integration of an existing neural network (NN) capable of providing information about the orientation of the cotton boll based on a predefined convention. Using this information, the NBV algorithm will generate potential next-step views to reduce uncertainty about the object's class while successfully planning a path that avoids obstacles and unwanted movements.

The manipulation aspect of the project will be implemented using the MoveIt library, which is part of the ROS 2 environment. Finally, the entire system will be integrated using behavior trees, leveraging the BehaviorTree.CPP library to ensure a more robust and modular solution.

The work is under the scope of the following projects:

• DEMETER 5.0: Sistema digitalizado para la producción sostenible de algodón hidropónico en invernadero (web)