Hierarchically depicting vehicle trajectory with stability in complex environments.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Science

Background:

• Autonomous robots offer societal and economic benefits but struggle with agile navigation in complex environments.
• Human pathfinding excels due to superior spatial awareness and experience utilization, unlike current robotic systems.
• Existing algorithms often degrade in performance as environmental complexity increases.

Purpose of the Study:

• To design a neural network simulating human intuitive pathfinding for robots.
• To integrate global environmental data and past experiences for feasible pathway identification.
• To enhance robot navigation efficiency and stability in complex, real-world scenarios.

Main Methods:

• A neural network was developed to mimic human spatial awareness and experience-based pathfinding.
• A numerically stable spatiotemporal trajectory optimizer with a bilayer polynomial representation was introduced.
• Differential flatness was leveraged for efficient optimization and singularity elimination in trajectory generation.
• A hierarchical motion planner combining path planning and trajectory refinement was implemented.

Main Results:

• The proposed neural network demonstrated stable computational performance, outperforming traditional algorithms in complex settings.
• The trajectory optimizer achieved robust convergence to continuous and feasible motion, even in challenging maneuvering situations.
• Large-scale maze experiments validated the hierarchical motion planner's robust and efficient navigation capabilities.

Conclusions:

• The developed planner enhances stable robot navigation in complex environments.
• This approach is expected to significantly advance the field of robotic autonomy.
• The integration of human-like spatial reasoning and advanced optimization offers a promising direction for future robotics research.