Accurate collection of plant phenotyping is critical to optimising sustainable farming practices in precision agriculture. Traditional phenotyping in controlled laboratory environments, while valuable, falls short in understanding plant growth under real-world conditions. Emerging sensor and digital technologies offer a promising approach for direct phenotyping of plants in farm environments. This study investigates a learning-based phenotyping method using neural implicit surfaces reconstruction to achieve accurate in situ phenotyping of pepper plants in greenhouse environments. To quantitatively evaluate the performance of this method, traditional point cloud registration on 3D scanning data is implemented for comparison. Experimental result shows that NIR (neural implicit surfaces reconstruction) achieves competitive accuracy compared to the 3D scanning method. The mean distance error between the scanner-based method and the NeRF (neural radiance fields)-based method is 0.811 mm. This study shows that the learning-based NeRF method has similar accuracy to the 3D scanning-based method but with greater scalability and faster deployment capabilities.

Unstructured orchard canopies present a major challenge for robotic fruit harvesting, where unpredictable branch contact and visually occluded foliage prevent reliable manipulation using vision alone. Rigid end‐effectors are vulnerable to this interaction, risking damage to gripper and canopy. Just as human pickers rely on physical compliance and tactile feedback to navigate dense canopies, this work presents a compact two‐degree‐of‐freedom backdrivable robotic wrist, the Transparent Robotic Agricultural Wrist (TRAW), designed to accommodate and estimate canopy interaction torques during apple harvesting. A capstan‐driven differential transmission achieves mechanical transparency through passive backdrivability, enabling open‐loop impedance control without passive compliant elements. A dynamic model incorporating gravity compensation enables real‐time sensorless interaction torque estimation from motor current. Benchtop experiments validate the stiffness rendering accuracy (RMS error 0.29 N m rad ⁻¹ across 2.00–20.00 N m rad ⁻¹ ) and torque tracking performance (RMS error 0.035 N m). Field trials in a commercial apple orchard evaluate wrist compliance during canopy approach and apple envelopment across 120 trials of varying difficulty, demonstrating that the TRAW safely accommodates canopy interaction while characterizing the interaction torques encountered. These results establish wrist‐level compliance with sensorless torque estimation as an effective approach for safe physical interaction in unstructured canopy environments.