Digital-Twin-Oriented Virtual Training Environment for Agricultural Robot Navigation: A Vineyard Rover Case Study.

Area of Science:

• Agricultural Robotics
• Digital Twin Technology
• Simulation-Based Training

Background:

• Virtual training environments offer advantages for agricultural robotics development.
• They provide a safe space for algorithm evaluation, reducing risks to robots and crops.
• Virtual environments enable efficient data generation under diverse conditions, improving model adaptability.

Purpose of the Study:

• To reinterpret digital twin technology as an active training environment for agricultural robots.
• To propose a layered framework for training agricultural robots in virtual settings.
• To highlight agriculture as a strong use case for virtual training due to field variability and experimentation costs.

Main Methods:

• Utilized digital twin technology for vineyard robot development.
• Incorporated sensor simulation, procedural map generation, and agriculture-specific movement models.
• Developed and validated a framework for training agricultural robots in virtual environments.

Main Results:

• An autonomous reinforcement learning agent was trained entirely in a virtual environment.
• The agent successfully navigated to 155 out of 161 target points in a simulated vineyard.
• The framework demonstrated the effectiveness of virtual training for agricultural robot navigation.

Conclusions:

• Digital twins serve as active training spaces, not just passive mirrors, for agricultural robots.
• The proposed layered framework facilitates robust training of agricultural robots in virtual settings.
• Virtual training environments are crucial for advancing agricultural robotics, addressing challenges like field variability and labor shortages.