CAREER: Beyond Visible: Autonomous Soft Agro-Bioproduct Manufacturing with Lifelong Hyperspectral-Guided Imitation Learning for Mass Personalization

The agri-food manufacturing sector plays a critical role in economic stability and food security, yet it remains highly dependent on manual labor due to the inherent variability of soft biological products. Differences in shape, texture, and material properties limit the effectiveness of conventional automation approaches that are successful in highly structured manufacturing environments. This Faculty Early Career Development Program (CAREER) project addresses these challenges by advancing autonomous manufacturing systems capable of handling soft agro-bioproducts with adaptability and precision. By enabling robots to perceive and respond to product texture without physical contact, the work supports autonomous food processing, improved product quality, and resilient domestic food supply chains. The project contributes to the national interest by strengthening manufacturing competitiveness and advancing the scientific foundations of automation in a sector essential to the public. Educational and workforce development activities will prepare students for emerging careers at the intersection of manufacturing, robotics, and agricultural engineering through integrated classroom and experiential learning, hands-on workshops with local primary, secondary, and community college students, and active engagement of industry stakeholders through instruction, field visits, and extension workshops. These efforts support workforce development and broaden the impact of advances in agri-food engineering, manufacturing systems, and robotics. The technical objective of this project is to establish a unified learning framework that enables autonomous robotic systems to interpret and utilize non-contact spectral information to guide the handling and processing of soft agro-bioproducts. The research pursues three integrated efforts. First, a dataset linking non-invasive spectral measurements with mechanical texture properties will be developed across multiple representative soft agro-bioproducts, establishing a transferable texture representation for manufacturing tasks. Second, an imitation learning framework will be created to translate human demonstration knowledge into robotic control policies by connecting visual observations and inferred texture information to multi-axis manipulation strategies. Third, a lifelong learning approach will be designed to allow the learned representations to generalize across previously unseen agro-bioproducts, enabling scalable and personalized manufacturing without extensive retraining. Together, these efforts advance the fundamental understanding of adaptable, sensing-guided robotic manufacturing systems for soft agro-bioproducts. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2542318 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Dongyi Wang | Institution: University of Arkansas Agricultural Experiment Station, FAYETTEVILLE, AR | Award Amount: $511,074 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2542318 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2542318.html