Simulation software and virtual environments for acceleration of agricultural robotics: Features highlights and performance comparison

Redmond R. Shamshiri, Ibrahim A. Hameed, Lenka Pitonakova, Cornelia Weltzien, Siva K. Balasundram, Ian J. Yule, Tony E. Grift, Girish ChowdharyInternational Journal of Agricultural and Biological Engineering 2018 · DOI

TL;DR

This paper provides a comprehensive review and performance benchmark of simulation platforms—such as V-REP, Gazebo, and ARGoS—specifically for agricultural robotics. By demonstrating a high-fidelity case study of a scouting robot in a virtual citrus orchard, the authors show how simulation can drastically reduce the cost and risk of developing autonomous systems for the "unstructured" and unpredictable environments of modern farms.

What Problem Does This Paper Address?

Developing agricultural robots is notoriously difficult. Unlike industrial robots that operate in controlled factory settings, ag-bots must navigate "unstructured" environments: uneven terrain, varying light conditions, and delicate, irregularly shaped plants.

The authors point out a frustrating gap: despite decades of research, very few agricultural robots (aside from milking systems) have reached commercial viability. Testing physical prototypes in the field is expensive, seasonally dependent, and carries a high risk of damaging both the robot and the crops. This paper argues that the bottleneck isn't just hardware, but the lack of a standardized, high-performance simulation framework where researchers can safely iterate on complex sensing and control algorithms.

Methods

The researchers took a two-pronged approach:

  1. A Comparative Review: They evaluated over nine simulation platforms (including Webots, Gazebo, RoboDK, and OpenRAVE) based on their physics engines, CAD support, and compatibility with the Robot Operating System (ROS).
  2. Quantitative Benchmarking: They performed "stress tests" on three leading simulators—V-REP (now known as CoppeliaSim), Gazebo, and ARGoS—using two scenarios: a "Small Scene" (simple 2D plane) and a "Large Scene" (a complex industrial model with 416,000 vertices). They measured:
  3. Real-time factor (R): How much faster or slower the simulation runs compared to real life.
  4. CPU/Memory Usage: The computational "weight" of the software.
  5. Proof-of-Concept Case Study: They built a custom virtual citrus orchard in V-REP, integrating it with ROS for data exchange and MATLAB for PID control and image processing.

Datasets

While this isn't a traditional machine learning paper using static datasets, the authors utilized specific virtual benchmarks:

  • Small/Large Scene Models: Standardized 3D environments used to measure simulator performance under varying vertex counts.
  • Virtual Citrus Orchard: A custom-built environment featuring high-fidelity 3D models of citrus trees and a prototype 4-wheel drive scouting robot equipped with a 6-DOF manipulator, Kinect sensors, and laser scanners.

Key Insights and Ideas

  • The "Swiss Army Knife" vs. The "Swarmer": The study highlights that there is no "one size fits all" simulator. V-REP is the most feature-rich and user-friendly for complex single-robot tasks (like fruit picking) because it allows for CAD editing directly in the scene. However, ARGoS is vastly superior for swarm robotics, as it can simulate thousands of simple robots by efficiently utilizing multi-threading.
  • Headless Performance: For heavy computational tasks, running simulators in "headless" mode (without a Graphical User Interface) significantly boosts the real-time factor, especially for Gazebo.
  • The Power of Middleware: By using ROS as a bridge, the authors demonstrate that the exact same control code used in a V-REP simulation can be deployed to a physical robot with minimal changes, effectively bridging the "reality gap."
  • Physics Engine Flexibility: V-REP’s ability to swap between four different physics engines (Bullet, ODE, Newton, Vortex) allows researchers to prioritize either speed or high-precision contact dynamics depending on the task (e.g., navigating a field vs. grasping a delicate sweet pepper).

Key Results

  • Speed: In GUI-based tests with 50 robots, ARGoS maintained a stable real-time factor, while V-REP struggled with the resource load.
  • Efficiency: Gazebo proved to be a strong middle ground, outperforming ARGoS in large-scale "headless" environments but requiring more Linux expertise to set up compared to V-REP.
  • Case Study Success: The virtual citrus orchard successfully simulated 360° environment scanning and autonomous navigation. The authors showed that they could tune PID controllers in MATLAB and see the immediate physical effects on the simulated robot’s joints and sensors without any risk of hardware breakage.

Limitations and Future Work

The authors acknowledge that while simulators are powerful, they cannot perfectly replicate the "noise" of the real world—such as unexpected wind gusts or sensor interference from dust.

Future Directions: * Standardization: The paper strongly advocates for an open-source software platform specifically tailored for agriculture, allowing research groups to share 3D plant models and "virtual fields." * Swarm Integration: Moving toward multi-robot and multi-drone systems that collaborate to optimize farming inputs (water, fertilizer, pesticides) in real-time. * Improved Meshes: There is a need for more efficient ways to handle highly complex 3D plant meshes without tanking simulation frame rates.

Summary updated

Abstract

Research efforts for development of agricultural robots that can effectively perform tedious field tasks have grown significantly in the past decade. Agricultural robots are complex systems that require interdisciplinary collaborations between different research groups for effective task delivery in unstructured crops and plants environments. With the exception of milking robots, the extensive research works that have been carried out in the past two decades for adaptation of robotics in agriculture have not yielded a commercial product to date. To accelerate this pace, simulation approach and evaluation methods in virtual environments can provide an affordable and reliable framework for experimenting with different sensing and acting mechanisms in order to verify the performance functionality of the robot in dynamic scenarios. This paper reviews several professional simulators and custom-built virtual environments that have been used for agricultural robotic applications. The key features and performance efficiency of three selected simulators were also compared. A simulation case study was demonstrated to highlight some of the powerful functionalities of the Virtual Robot Experimentation Platform. Details of the objects and scenes were presented as the proof-of-concept for using a completely simulated robotic platform and sensing systems in a virtual citrus orchard. It was shown that the simulated workspace can provide a configurable and modular prototype robotic system that is capable of adapting to several field conditions and tasks through easy testing and debugging of control algorithms with zero damage risk to the real robot and to the actual equipment. This review suggests that an open-source software platform for agricultural robotics will significantly accelerate effective collaborations between different research groups for sharing existing workspaces, algorithms, and reusing the materials. Keywords: agricultural robotics, precision agriculture, virtual orchards, digital agriculture, simulation software, multi-robots DOI: 10.25165/j.ijabe.20181104.4032 Citation: Shamshiri R R, Hameed I A, Pitonakova L, Weltzien C, Balasundram S K, Yule I J, et al. Simulation software and virtual environments for acceleration of agricultural robotics: Features highlights and performance comparison. Int J Agric & Biol Eng, 2018; 11(4): 15–31.
BibTeX
@article{shamshiri2018simulation,
  author = {Redmond R. Shamshiri and Ibrahim A. Hameed and Lenka Pitonakova and Cornelia Weltzien and Siva K. Balasundram and Ian J. Yule and Tony E. Grift and Girish Chowdhary},
  title = {{Simulation software and virtual environments for acceleration of agricultural robotics: Features highlights and performance comparison}},
  journal = {International Journal of Agricultural and Biological Engineering},
  year = {2018},
  doi = {10.25165/j.ijabe.20181103.4032},
}
Figure