A method includes generating an image, segmenting the image to identify a component of an agricultural item, detecting one or more agricultural features of the agricultural item based on the image, the one or more agricultural features being associated with the component of the agricultural item, generating a two-dimensional grab-point based on the component of the agricultural item and the one or more agricultural features, and generating a three-dimensional grab-point based on the two-dimensional grab-point and a depth estimation of the agricultural item.