Crowd navigation in a multi-room environment: a model predictive control framework for mobile robots

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Mobile robots require safe navigation in human-populated, non-convex environments.
• Collision-free motion planning is crucial for robot operation in crowded spaces.

Purpose of the Study:

• To develop a sensor-based model predictive control (MPC) scheme for safe crowd navigation.
• To improve robot safety and efficiency in complex, multi-room environments.

Main Methods:

• Decomposing free space into convex regions to create a topological graph for high-level planning.
• Fusing 2D LiDAR and RGB-D camera data with Kalman filters (KFs) for robust human motion estimation and prediction.
• Integrating human motion predictions into an MPC controller with discrete-time control barrier functions (DT-CBF) for collision avoidance.

Main Results:

• The proposed framework successfully navigates cluttered, multi-room scenarios.
• Significant improvements in collision avoidance and task success rates were observed.
• Validation through high-fidelity simulations and real-world experiments with the TIAGo mobile manipulator.

Conclusions:

• Integrating vision-based semantic data with geometric constraints enhances robot safety in crowded environments.
• The developed MPC scheme provides effective collision-free motion planning for mobile robots.
• The system demonstrates robust performance in complex, dynamic human-populated settings.