Advanced motion planning for real-time and scalable autonomous systems in high-dimensional environments

Abstract

Enabling safe, scalable, and intelligent autonomy in complex environments remains central to nextgeneration robotics. Autonomous vehicles, surgical robots, and manipulators demand real-time motion planning resilient to high-dimensional complexity. Traditional algorithms struggle under kinodynamic constraints, motivating hybrid learning–planning architectures. Integrating deep reinforcement learning with sampling-based planners enhances exploration efficiency, while hierarchical control improves task decomposition. A checkpoint-based rewarding mechanism addresses internal reward contradictions, and Q-value–guided exploration ensures adaptive sampling. Anchored in Trustworthy AI and Operational Design Domain principles, this hybrid approach outperforms trial-and-error baselines and advances robust, explainable, and resource-efficient autonomy across dynamic, safety-critical robotic applications.