This project focuses on minimizing the latency in agricultural decision-making by harnessing generative AI, mobile robotics, and edge AI. The long-term goal of the proposed project is to enable real-time adaptation of the data collection processes over multiple types of crops and minimize the time at which collected data can be processed to make decisions in support of critical agronomic tasks. To this end, the objectives of this project include 1) generative AI and robotics for scalable multi-modal data collections, and 2) edge AI for on-demand data analysis and real-time decision-making, as well as a concurrent, multi-stage evaluation of the foundational and applied research methods studied in this project. We consider medium-scale (i.e. a few kg of weight) unmanned aerial vehicles (UAVs) tasked to operate both over and inside/under the canopy in citrus groves and vineyards. The specific agronomic tasks considered include optimal harvesting time identification for grapes, efficient nutrient management for citrus, and soil health monitoring in both vineyards and citrus groves, over distinct climatic zones ([Southern] California, Florida, and [Northern] Greece).
The project investigates the deployment of pervasive, intelligent, and autonomous agricultural robotics at the frontier of the farming workforce and agricultural robotics and automation technology by creating new, expanded, and unique user-centered frameworks. The project uniquely innovates along five fundamental agricultural robotics and automation technology and agricultural workforce research directions. 1) Novel principles to co-design actuation and perception for safe, reliable, and efficient robotic harvesters. 2) Effective machine vision mechanisms to understand farmworker activities in harvesting. 3) Efficient robot planning techniques cognizant of human activities. 4) Participatory design approach for precision farming technology trust and adoption. 5) Advancement of human-robot multitasking toward sustainable agriculture. The project actively engages stakeholders (farmworkers, farm owners, packing house specialists) to assess current standards and practices and then integrate feedback after in-field demonstrations to inform iterative modifications of devices and systems.
The project develops an attention-driven software architecture that can identify and prioritize critical information from sensors, enabling timely decision-making while considering resource constraints and uncertainties in the environment. This architecture holistically optimizes computation scheduling, perception, and planning by adapting to the context and anticipating future actions. Three important advances to be made in this project are (1) context adaptive scheduling of autonomous computation pipelines, (2) learning-based perception to anticipate future actions in dynamic environments, and (3) motion planning and decision-making based on anticipated actions in the presence of uncertainty. By interrelating these components through the attention-driven architecture, this project tackles fundamental challenges associated with time-sensitive scenarios in resource-constrained autonomous systems.
Reintroduction of insect decomposers into agricultural systems can alleviate waste removal costs, rapidly generate usable products (fertilizer), and lead to new revenue streams (insect biomass). The Black Soldier Fly (BSF) is an effective decomposer of any type of organic matter and is used worldwide for waste management, production of fertilizer (frass), and as feed for livestock, aquaculture, and pets. However, BSF rearing by the standard "batch" method (carefully measured cohorts of larvae with set quantities of waste) currently requires substantial human labor, which hinders on-farm use. An alternative "steady-state" rearing system involves continuous rearing of BSF in bioreactors and is less labor intensive, but still requires some human maintenance. These maintenance tasks are targets for introducing cost-conscious sensing and automation. The goal of this seed grant project is to engineer, build, and test a prototype automated steady-state system suitable for on-farm use. By automating on-farm BSF rearing, we will expand the utilization of waste in agricultural systems and engineer new products that utilize materials from agriculture.
The project investigates real-time, sensor-based closed-loop control for soft wearable robotics targeted to serve as upper extremity (UE) assistive devices. The project advances fundamental engineering knowledge in three ways. 1) Visual sensing for biomechanical applications via lensless cameras, that can take arbitrary shapes and have an ultra-thin form factor that allows seamless integration with soft materials and surfaces. The project develops a new class of visual object recognition and tracking algorithms for the use of lensless camera systems in UE assistive devices. 2) Soft fluidic logic circuits to reduce the amount of rigid electronic components required to control and operate pneumatically actuated soft robots. This technique is valuable in assistive technology applications where physical interaction with humans is better achieved if the size and weight of the device components remain low. 3) Shared human-robot admittance control, that uses data in real-time to estimate dynamic models for the device and its interaction with the human. The goal is to provide as-needed assistance by autonomously switching between two operating modes: either yielding to the user's intention or applying assistive forces to help the user's arm reach the desired object. Overall, this research will create new forms of embodied intelligence to advance the capacity of wearable assistive soft robots to perceive and act in collaboration with a human user.
The project investigates how compliance embedded into a legged robot can be harnessed to facilitate control and computation, with an eye to enabling efficient and resilient navigation in real agricultural fields. Research activities innovate along three key foundational robotics research directions. 1) Hardware design and dynamic modeling: The project offers fundamental insights and develops models regarding the effect of various forms of compliance on center of mass motion and gait stabilization for certain classes of legged robots and introduces new hardware designs that can harness compliance and enable principles of morphological computation. 2) Locomotion control: The project establishes compliance-aware legged locomotion controllers according to principles of whole-body and central pattern generator-based control to enable efficient closed-loop legged locomotion over a range of engineered and natural unstructured terrains. 3) Non-holonomic motion planning and autonomous navigation: The project develops non-holonomic motion planners that rely upon and utilize distinctive features of robot body morphology and embedded compliance for efficiency and resilience during autonomous legged locomotion over real agricultural fields. This research can transform the science and technology of autonomous legged robots by making them more efficient and resilient in their operation, and thus unlock legged robots' full potential in precision agriculture.
Funded in 2021 by the University of California Office of the President Multi-Campus Research Program Initiative, the Labor and Automation in California Agriculture (LACA) assembles interdisciplinary researchers to solve food system resilience and human-centered agriculture technology to transform the workforce and environment of the future. This MRPI includes researchers from the University of California Merced, Berkeley, Davis, and Riverside. The LACA team is focused on four research thrusts; AgTech: developing cutting-edge technology systems to monitor and manage resources and applications; Labor Impacts: engaging socio-economic factors for incorporation of new technologies; Environmental Impacts: create an agricultural model that incorporates data from AgTech for more sustainable farming practices; and Underlying and Emerging Problems: deals with policy and labor related issues that may arise during the course of the LACA research. The AgTech thrust involves developing and testing novel Agricultural technology systems, specifically three types of stationary and robotic systems: 1) Multi-modal Precision Irrigation: An irrigation system that incorporates water resource management; 2) Remote Sample Acquisition and Analysis: A scalable robotics design and prototype that can accurately assess and identify disease and pests while supporting environmental decision-making on fertilizer and pesticide application; and 3) Decision Making for Shared Autonomy in the Field: An autonomous decision-making system that can support agricultural task assignment decisions.
This project seeks to develop and deploy heterogeneous teams of autonomous robots (specifically, aerial and ground robots) to enable frequent and dense sampling in the field. The motivating hypothesis is that increased sampling density and frequency can indicate noticeable spatiotemporal variability in water potential that would remain undetected because of insufficient sampling resolution. To this end, the project incorporates 1) the development of a robotized pressure chamber, 2) visual sensing for accurate determination of leaf water potential, 3) multi-robot coordination and planning, and 4) extensive field testing and evaluation across multiple crop species.