Online Multi-Contact Motion Replanning for Humanoid Robots with Semantic 3D Voxel Mapping: ExOctomap

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Humanoid robot navigation in unknown environments requires efficient online mapping and motion planning.
• Traditional methods using polygon meshes or primitive extraction face challenges in complex, dynamic settings.

Purpose of the Study:

• To develop a rapid motion-replanning technique for humanoid robots.
• To enable autonomous navigation in unknown territories using semantic 3D voxel mapping.

Main Methods:

• Introduced an innovative Extended-Octomap (ExOctomap) for semantic 3D voxel mapping.
• Utilized Octree structure to store environmental normal vectors, Euclidean Cluster Extraction, and principal component analysis.
• Implemented 3D grid sampling for efficient 6D contact pose search and collision-free trajectory planning.

Main Results:

• Achieved O(log N) efficiency in semantic accessibility from position queries.
• Demonstrated efficient adaptation of multi-contact motions across diverse environments.
• Obtained near real-time planning speeds between 13.8 ms and 115.7 ms per contact.

Conclusions:

• The proposed semantic voxel mapping and motion-replanning framework significantly enhances humanoid robot autonomy.
• The Extended-Octomap provides an efficient and scalable solution for real-time environmental sensing and navigation.