A Self-Collision Detection Algorithm of a Dual-Manipulator System Based on GJK and Deep Learning

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Safe operation of multi-manipulator systems requires robust self-collision detection, especially in dynamic environments.
• Current methods struggle to balance detection efficiency and accuracy simultaneously.
• Integrating artificial intelligence offers a promising avenue for improving detection capabilities.

Purpose of the Study:

• To develop a novel two-level self-collision detection algorithm (DLGJK) for dual-manipulator systems.
• To enhance detection efficiency and accuracy using artificial intelligence and the GJK algorithm.
• To address real-time self-collision detection challenges in fast-continuous and high-precision robotic applications.

Main Methods:

• Generated a dataset and trained a deep neural network (DLNet) based on the Gilbert-Johnson-Keerthi (GJK) algorithm.
• Proposed the DLGJK algorithm, a two-level approach combining DLNet for initial risk assessment and GJK for precise detection.
• Implemented DLNet to filter non-risky states, reducing unnecessary computations.

Main Results:

• The DLGJK algorithm achieved a 97.7% reduction in the expected time for single self-collision detection within the workspace.
• DLNet effectively identified states with no self-collision risk, significantly improving overall detection efficiency.
• The GJK algorithm provided accurate fine detection for states identified as potentially risky by DLNet.

Conclusions:

• The DLGJK algorithm offers a significant improvement in self-collision detection efficiency and accuracy for dual-manipulator systems.
• This AI-driven approach reduces system overhead and enhances performance in path planning.
• The algorithm demonstrates scalability for broader multi-manipulator system applications.