Trajectory Prediction-Enabled Self-Decision-Making for Autonomous Cleaning Robots in Semi-Structured Dynamic Campus...

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Autonomous cleaning robots operate in complex, dynamic environments with diverse agents like pedestrians and vehicles.
• Safe navigation requires anticipating surrounding agents' future movements beyond simple reactive obstacle avoidance.

Purpose of the Study:

• To develop and evaluate a trajectory prediction-enabled self-decision-making framework for autonomous cleaning robots in campus environments.
• To enhance robot decision-making by incorporating predictive models of agent behavior.

Main Methods:

• A learning-based multi-agent trajectory prediction model was trained offline using benchmark and real-world data.
• Predicted trajectories were integrated as priors into the robot's online decision-making and planning.
• Evaluation was conducted on a high-fidelity simulation platform using data-driven scenario reconstruction.

Main Results:

• The framework demonstrated improved motion stability and fewer abrupt adjustments in interaction scenarios.
• Short-term prediction horizons achieved over 90% ADERate and FDERate.
• Lane-change prediction accuracy reached approximately 79%, with stable speed tracking in medium-density traffic.

Conclusions:

• Integrating trajectory prediction significantly enhances the stability and predictability of autonomous cleaning robot behavior.
• The proposed framework offers a robust solution for safe and efficient robot operation in interaction-rich environments.