Probabilistic Sampling Networks for Hybrid Structure Planning in Semi-Structured Environments

Area of Science:

• Robotics
• Artificial Intelligence
• Manufacturing Engineering

Background:

• Traditional motion planning methods struggle in high-dimensional spaces, limiting adaptable industrial robots.
• Intelligent manufacturing requires efficient and resilient motion planning solutions.

Purpose of the Study:

• To develop a novel hybrid motion planner for adaptable industrial robots.
• To enhance motion planning performance in high-dimensional spaces using Dempster-Shafer evidence theory.

Main Methods:

• A hybrid motion planner integrating a probabilistic sampling network (PSNet) and an enhanced artificial potential field (EAPF).
• PSNet, comprising motion planning (MPM) and fusion sampling modules (FSM), generates multimodal path distributions.
• FSM uses Gaussian resampling for collision correction, while EAPF repairs local paths in worst-case scenarios.

Main Results:

• The proposed method quickly finds directly connectable paths in diverse environments.
• The system reliably avoids sudden obstacles, demonstrating enhanced resilience and adaptability.
• Experimental results validate the effectiveness of the collaborative motion planning strategy.

Conclusions:

• The Dempster-Shafer evidence theory-based hybrid motion planner significantly improves robotic motion planning efficiency.
• The integration of PSNet and EAPF offers a robust solution for adaptable industrial robots in intelligent manufacturing.
• The developed system enhances path planning reliability and adaptability in dynamic environments.