A Unified Local Risk Map for Uncertainty-Aware Mobile Robot Navigation in Cluttered and Dynamic Environments.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Mobile robot navigation in cluttered and dynamic environments faces challenges due to perception and actuation uncertainties.
• Traditional methods like fixed safety margins or costmap inflation layers are difficult to tune and may lead to unsafe or inefficient navigation.

Purpose of the Study:

• To develop a unified risk-cost map that integrates multiple sources of uncertainty for improved robot navigation.
• To enhance the robustness and safety of mobile robots in complex environments.

Main Methods:

• A probability-inspired risk-cost map was developed, integrating perception and actuation uncertainty, dynamic obstacle prediction, and occlusion-aware memory.
• A local path-modification module adapted global paths using the risk map, interfacing with a Model Predictive Path Integral (MPPI) controller.
• The framework was validated in Gazebo simulations with varying uncertainty conditions and obstacle types.

Main Results:

• The proposed method demonstrated superior robustness compared to conventional costmap-based baselines.
• Fewer aborted goals were observed in cluttered environments.
• A substantial reduction in collision events was achieved, particularly in the presence of dynamic obstacles.

Conclusions:

• The unified risk-cost map provides a more effective approach to mobile robot navigation in uncertain and dynamic environments.
• The proposed method enhances safety and efficiency, outperforming traditional navigation techniques.