Agibot: An Open-Source Modular Swerve-Drive Robot Platform for Agricultural Robotics Research and Education

Moeen Ul Islam, Bishal Adhikari, Kori Edwards, Lorien Harvey, J. Alex Thomasson, Dong ChenarXiv 2026 · arXiv

TL;DR

Agibot is an open-source, low-cost, and modular robotic platform designed to lower the barrier to entry for agricultural robotics research and education. It features a highly agile four-wheel independent swerve drive and a robust "sense-plan-act-verify" autonomy framework that ensures tasks like seed dispensing and plant manipulation are not just executed, but successfully completed in variable field-like conditions.

What Problem Does This Paper Address?

Developing autonomous systems for agriculture is notoriously difficult. Unlike the structured environments of factories, farms present irregular terrain, shifting lighting, and biological variability. Currently, researchers and educators face a stark choice: use expensive, proprietary commercial platforms that are difficult to modify, or rely on low-cost educational robots that are typically designed for flat, indoor floors and lack the power for complex agricultural tasks.

The authors introduce Agibot to fill this gap. It provides a "middle ground" platform that is affordable (roughly $760–$1,680 depending on the compute module), fully open-source, and capable of handling the unstructured nature of agricultural environments.

Methods

Agibot’s design is built on three pillars: modular hardware, hierarchical computing, and a closed-loop autonomy framework.

  • Mechanical & Drive System: The robot uses a 3D-printed chassis reinforced with steel rods. Its standout feature is a four-wheel independent swerve drive (8 degrees of freedom). This allows the robot to "strafe" sideways or rotate in place, which is critical for navigating tight crop rows. It also features a hybrid 3-DOF manipulator capable of both contact-based tasks (like toppling plants) and precision dispensing (like planting seeds).
  • Hierarchical Architecture: As shown in the electronics overview, the system separates high-level "thinking" from low-level "reflexes." Fig.

  • Hierarchical electronics architecture of Agibot. A high-level computer (e.g., NVIDIA Jetson AGX Orin) handles vision and planning, while two ESP32 microcontrollers manage real-time motor control and sensor data acquisition.

  • Sense-Plan-Act-Verify Loop: Most robots follow a "sense-plan-act" cycle. Agibot adds a crucial Verify step. After performing an action, the robot uses its cameras and sensors to confirm the physical outcome (e.g., "Did the seed actually land in the container?"). If not, it triggers a recovery or retry behavior.

  • Perception: The "RailNav" pipeline uses instance segmentation (YOLOv8n-seg) to identify crop rows and calculate lateral and heading errors for navigation.

![Fig.

  1. Hierarchical electronics architecture of Agibot.](/figures/art_e7b7fc88e0904499bde5134287b53b28)

Datasets

The platform was evaluated in a custom-built, representative agricultural testbed featuring parallel navigation rails, soil-covered terrain, and 3D-printed plant targets.

  • Perception Evaluation: The RailNav pipeline was tested on a dataset of 68 held-out images captured under various lighting and soil conditions.
  • Cross-Platform Benchmarking: The authors benchmarked their perception models across four different hardware configurations: NVIDIA Jetson AGX Orin, Orin Nano, Raspberry Pi 5 with a Hailo-8 accelerator, and a standalone Raspberry Pi 5.

Key Insights and Ideas

  • The Power of Verification: The "Verify" step is the platform's secret sauce. In integrated trials, adding post-action verification increased the manipulation success rate from 96.8% to 99.1%. This demonstrates that for agricultural tasks, closing the loop on the outcome is just as important as closing the loop on the motion.
  • Swerve Drive Agility: The independent swerve modules (Fig. 2a) allow for simultaneous lateral and heading corrections. This "crab-walk" capability is far more efficient for row-crop navigation than traditional differential or Ackermann steering. Fig.

  • (a) A single swerve module.

  • Compute Modularity: By using a standardized interface, the robot can be powered by anything from a $60 Raspberry Pi to a $1,000 Jetson. This allows users to scale the robot's "intelligence" to match their budget and task complexity.

![Fig.

  1. (a) A single swerve module.](/figures/art_710b290740df4e18a54d4e449b1cbf1b)

Key Results

  • Navigation Reliability: Agibot achieved a 98% trial success rate across five different environmental conditions (low/high light, no soil, thin/heavy soil).
  • Task Success: In independent tests, the robot achieved a 100% success rate for plant toppling (43/43) and a 94.5% success rate for seed dispensing (52/55).
  • Latency: The full navigation pipeline (inference to control) runs at approximately 159ms on the Jetson AGX Orin, which is sufficient for the robot's operating speeds in the testbed.
  • Control Precision: Tuning studies showed that a "nominal" gain setting provided the best balance, achieving a lateral Root Mean Square Error (RMSE) of about 42.92 pixels while maintaining 100% course completion.

Limitations and Future Work

While Agibot performed excellently in a controlled testbed, the authors acknowledge that field-scale reliability in real-world farms remains to be proven. The current 3D-printed components and small-scale wheels may need ruggedization for long-term outdoor use.

Future work will focus on testing Agibot in actual field conditions, integrating more diverse sensors (like LiDAR), and developing new end-effectors for a wider range of agricultural operations. The authors also plan to integrate the platform into university robotics curricula through dedicated lab modules.

Summary updated

Abstract

Agricultural production faces persistent labor shortages and increasing demand for automation, yet progress in agricultural robotics remains constrained by the limited availability of affordable, flexible, and accessible research platforms. This paper presents Agibot, an open-source, modular, low-cost agricultural robotics platform designed for autonomy research and experimentation, with potential applications in hands-on robotics education. The platform integrates four-wheel independent swerve drive, swappable computing, distributed embedded control, multimodal sensing, and configurable manipulation within a unified sense-plan-act-verify framework supporting perception, control, navigation, manipulation, and autonomous task verification. The platform was evaluated through representative row-crop navigation and manipulation tasks under varying lighting and soil conditions. Experiments demonstrated reliable lane following, autonomous row-to-row navigation, target identification, seed dispensing, plant manipulation, and closed-loop verification of task success. Its modular architecture supports independent testing, modification, and validation of autonomy components for rapid prototyping and experimentation. These proof-of-concept tasks, controller-tuning studies, and integrated trials demonstrate the platform's research capabilities, while its accessible and reconfigurable design supports its use for hands-on robotics education. By combining affordability, modularity, open-source accessibility, configurable manipulation, and end-to-end sense-plan-act-verify autonomy, Agibot provides a practical platform for agricultural robotics research, technology development, and workforce training.
BibTeX
@misc{islam2026agibot,
  author = {Moeen Ul Islam and Bishal Adhikari and Kori Edwards and Lorien Harvey and J. Alex Thomasson and Dong Chen},
  title = {{Agibot: An Open-Source Modular Swerve-Drive Robot Platform for Agricultural Robotics Research and Education}},
  year = {2026},
  eprint = {2611.10235},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.10235},
}
Figure