MODNet: A Monocular Object-Based Depth Estimation Network for AI Robotic Chemists

Area of Science:

• Artificial Intelligence in Chemistry
• Robotics and Automation
• Computer Vision

Background:

• Machine learning and computer vision are transforming chemistry, but intelligent robotic arm operation is crucial for AI chemist platforms.
• Current AI chemist platforms require advanced sensory input and precise manipulation capabilities for complex tasks.

Purpose of the Study:

• To develop a machine vision-assisted system for intelligent robotic arm operations in AI chemists.
• To create a robust model for accurate object detection and distance estimation of chemical apparatus.
• To enable real-time guidance for robotic arm grasping in chemical laboratory settings.

Main Methods:

• Proposed MODNet, a chemical apparatus object detection and distance estimation model based on the CViG_II dataset.
• Integrated a monocular distance measurement method for real-time target distance detection.
• Validated the system using unit operations in the Spiro [fluorene-9,9'-xanthene] (SFX) synthesis process.

Main Results:

• MODNet achieved 95.8% experimental accuracy in detecting chemical apparatus, surpassing the original YOLOv8 algorithm.
• The integrated system demonstrated a distance measurement error margin of less than 5% and a real-time inference frame rate over 60 fps.
• Successfully provided real-time guidance for robotic arm grasping in chemical reagent preparation procedures.

Conclusions:

• MODNet offers a high-precision solution for object and distance detection of chemical apparatus, crucial for AI chemists.
• The vision-assisted system enables rapid, accurate robotic arm operations in laboratory environments.
• This work provides a foundational step towards fully autonomous chemical synthesis driven by AI robotic chemists.