Hefty: A Modular Reconfigurable Robot for Advancing Robot Manipulation in Agriculture

TL;DR

Hefty introduces a modular, reconfigurable mobile manipulation platform for agriculture, built on the Farm-ng Amiga base to enable rapid adaptation across mobility, sensing, power, computing, and fixture mounting. Through five configurations, the work demonstrates navigation and manipulation tasks including insect scouting, sensor insertion, pepper harvesting, and lantern-fly control, highlighting how modularity reduces capital costs and accelerates technology transfer. The platform relies on ROS for software integration, a high-performance onboard computing stack, and a diverse sensor payload that can be reconfigured for task-specific needs, with a dual-power system to support both mobility and complex sensing/manipulation. Limitations include compute power constraints and software environment management, with future work focusing on multi-arm capabilities, AI-driven configuration optimization, and virtualization-based workflows to enable broader, parallel research and end-user deployment.

Abstract

This paper presents a modular, reconfigurable robot platform for robot manipulation in agriculture. While robot manipulation promises great advancements in automating challenging, complex tasks that are currently best left to humans, it is also an expensive capital investment for researchers and users because it demands significantly varying robot configurations depending on the task. Modular robots provide a way to obtain multiple configurations and reduce costs by enabling incremental acquisition of only the necessary modules. The robot we present, Hefty, is designed to be modular and reconfigurable. It is designed for both researchers and end-users as a means to improve technology transfer from research to real-world application. This paper provides a detailed design and integration process, outlining the critical design decisions that enable modularity in the mobility of the robot as well as its sensor payload, power systems, computing, and fixture mounting. We demonstrate the utility of the robot by presenting five configurations used in multiple real-world agricultural robotics applications.

Figures (11)

• Figure 1: A complete view of Hefty in (a) also shows the keyboard, mouse, and monitor for interfacing with the system computer. The black mast carries the GPS and IMU sensors, and the aluminum has the perception sensors: a 3.3MP RGB camera, a PoE stereo camera, a Realsense camera, and a Lidar. Subfigures (b)--(e) showcase the various manipulator configurations employed in (b) pepper harvesting, incorporating a modified structure to elevate the mounting plate; (c) cornstalk sensor insertion; (d) soybean insect scouting; and (e) lantern-fly egg mass decimation.
• Figure 2: Default Amiga robot configuration, as provided by Farm-ng, before all the modifications to accommodate robot manipulation configurations.
• Figure 3: The crossing beams and the single beams (shown in (a) and (b), respectively) make the Amiga's structural joinery robust and modular. Images (c) and (d) show how the same joinery is applied to mechanical risers to increase the height clearance.
• Figure 4: A closer look at the modular design of the manipulator mounting plate used in Fig. \ref{['fig:tasks']} configuration examples. From the right, the images show the top, side and bottom orthographic views. The key modularity feature is using the single-bar clamps to mount to any 1.5" parallel bars.
• Figure 5: Examples of how 1.5in 80-20 Aluminum improves the modularity and reconfigurability of fixture mounting on the Hefty robot. From left: 3D models of a RealSense Camera (d455), an Oak-D W Pro camera below a Velodyne V16 Lidar, a SwiftNav GPS unit, and a weather-proof electric box housing the system computer and electrical system.