Vision–Language–Action Model for Agricultural Fresh Produce Handling with a Humanoid Robot
Year: 2026
Type: article
Abstract: Reaching, grasping, and placing fresh produce is among the most labor-intensive operations in agricultural production, yet most robotic solutions remain task-specific, single-crop systems that generalize poorly across the morphological diversity of fresh produce. This study investigates whether a large, pre-trained Vision-Language-Action (VLA) model can be adapted, with only modest demonstration data, to serve as a flexible language-conditioned produce-handling policy on a humanoid robotic platform. The NVIDIA GR00T N1.7-3B foundation model is fine-tuned for a standard 23-degree-of-freedom Unitree G1 humanoid robot fitted with two Dex3-1 threefinger dexterous hands, whose arm and hand joint streams form a 24-dimensional combined manipulation command space. Demonstrations were collected through a Meta Quest 3 extendedreality teleoperation pipeline, extended here with a language-labeled teleoperation recording mode that embeds a natural-language goal in each episode, yielding 605 language-conditioned demonstrations (261,724 synchronized frames) across four produce categories: lemon, peach, pear, and tomato. A dual-camera design paired a wide-angle head camera for global scene context with a wrist-mounted camera for close-range grasp feedback. The results of this study show that the fine-tuned policy achieved a 96.25% single-category trial success rate and 85% on mixed-scene trials requiring retrieval of the language-specified item from four simultaneously present categories. In addition, incorporating the wrist-mounted camera substantially improved the robot's ability to complete the placement phase with greater reliability and precision. This result shows that close-range wrist perception is a critical component of the dual-camera system, enabling robust pick-and-place performance rather than providing only a minor improvement. Together, these results demonstrate that VLA fine-tuning can transform a pre-trained humanoid foundation model into a reliable, data-efficient policy for multi-category fresh produce handling. This work provides a practical step toward flexible humanoid robotic systems for agricultural pick-and-place and sorting tasks in postharvest automation.
Source: AgriEngineering
Authors: Kapalik Khanal, Biswash Khatiwada, Stephen Afrifa, Sawyer Exum, Fan Wu, Ramesh Bahadur Bist
Institutions: Tribhuvan University, North Carolina State University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: gold
Estimated APC paid: $1,930
Funder: North Carolina State University
OpenAlex ID: W7172401530