A Survey on Learning Motion Planning and Control for Mobile Robots: Toward Embodied Intelligence

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Mobile robots are crucial for environmental monitoring and search-and-rescue.
• Embodied artificial intelligence (AI) enables robots to learn and evolve through environmental interaction.
• Motion planning and control are essential for effective robot operation.

Purpose of the Study:

• To provide a comprehensive overview and classification of learning-based motion planning and control for mobile robots.
• To categorize existing approaches based on system architectures and machine learning algorithms.
• To highlight key technologies and challenges in achieving embodied AI.

Main Methods:

• Categorization of approaches into five system architectures based on ML module application.
• Review of reinforcement learning (RL), imitation learning, and ML-MPC integration.
• Presentation of research on safe learning control, Sim2Real transfer, and large language models (LLMs).

Main Results:

• A taxonomy of learning-based motion planning from a system architecture perspective.
• Detailed review of ML techniques applied across different architectures.
• Identification of critical issues and challenges in embodied AI.

Conclusions:

• The survey offers a novel system architecture-based classification of embodied AI research in mobile robotics.
• It emphasizes technologies crucial for real-world applications.
• The work aims to advance embodied AI in mobile robots through future research directions.