Bridge Points Guided Neural Motion Planning in Complex Environments with Narrow Passages

Area of Science:

• Robotics
• Artificial Intelligence
• Motion Planning

Background:

• Intelligent robotic systems require robust motion and path planning for navigation in complex environments.
• Traditional sampling-based planners struggle with narrow passages, while learning-based methods can yield infeasible paths.
• A significant challenge lies in generating collision-free trajectories efficiently, especially in obstacle-rich configuration spaces (C-spaces).

Purpose of the Study:

• To develop a hybrid path planning framework that overcomes the limitations of existing methods in narrow passage scenarios.
• To improve the success rate and reduce planning time for robotic navigation in challenging environments.
• To combine the strengths of sampling-based and learning-based approaches for more reliable and efficient path generation.

Main Methods:

• A hybrid framework integrating improved sampling, structural abstraction, and neural prediction.
• A modified bridge-test sampler with directional perturbations and corridor checks for narrow passage sampling.
• Clustering of samples into representative bridge points forming a global graph, queried using greedy heuristic search and a neural local segment generator.

Main Results:

• The proposed method significantly outperforms classical and learning-based baselines in success rate and planning time.
• Validation across 2D maze maps, 3D voxel environments, and a complex 12-DOF manipulator task demonstrated superior performance.
• The framework retains a generalized form of probabilistic completeness with an optional repair module.

Conclusions:

• The hybrid framework offers a significant advancement in motion planning, particularly for environments dominated by narrow passages.
• This approach effectively balances planning efficiency and path reliability, addressing key limitations of prior methods.
• The method provides a robust solution for intelligent robotic navigation in complex, cluttered C-spaces.