Context-aware techniques for real-time decision making in autonomous mobile robots (Situation awareness Part A)

• Main Author: Parittotog, Apichaya
• Other Authors: Andy Khong W H
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Reinforcement learning; Robotics; Artificial intelligence
• Online Access: https://hdl.handle.net/10356/176279
• Institution: Nanyang Technological University

This final year project bridges the fields of Artificial Intelligence (AI) and robotics, aiming to establish an advanced framework for training autonomous robots through simulation. By leveraging the combined strengths of Autodesk ReCap Photo, NVIDIA Isaac Sim, and Stable Baselines, the initiative introduces a novel strategy for enhancing robots’ navigation and obstacle-avoidance capabilities using sophisticated reinforcement learning techniques. The fundamental approach involves constructing detailed 3D environments that replicate the complexities of real-life settings, providing a realistic and engaging context for training. A standout feature of this project is the integration of LiDAR sensors into the simulation framework, significantly enhancing the robots’ ability to perceive and interact with their environment dynamically. Furthermore, the introduction of a differential controller, implemented through Isaac Sim’s Action Graph, markedly improves the accuracy and realism of the robots’ movements, an essential factor for their successful application in varied real- world situations. Compared to a standard tutorial-based training environment, this project demonstrates substantial progress in scene adaptability, environmental understanding, navigation techniques, and goal and obstacle management. These improvements result in a more flexible and comprehensive training platform, equipping robots with the necessary skills to adeptly handle a wide array of situations and challenges. In essence, this project not only makes significant contributions to the domains of AI and robotics but also paves the way for future research into simulation-based robot training. It highlights the profound impact of integrating cutting-edge technologies to prepare autonomous robots for the intricate demands of real-world operations, providing a solid foundation for ongoing research and development of intelligent, versatile robotic systems.