Scene Interaction-Aware Path Planning for Mobile Robots With Planar Interaction Capabilities
By actively interacting with surroundings and modifying environmental conditions, mobile robots with planar interaction capabilities can achieve comprehensive autonomous planning in complex, multi-connected domain scenarios. This study introduces a scene interaction-aware planning framework that empowers agents to reshape their environment and resolve navigation challenges across disconnected domains. The framework characterizes bidirectional agent-environment relationships and decomposes the cross-domain planning problem into three interconnected sub-problems: domain-transition sequencing, interaction process planning, and static path search. Correspondingly, a hierarchical multi-layer graph search algorithm is proposed, where higher-level strategic searches utilize lower-level feasibility results while providing global guidance. Theoretical analysis demonstrates that this hierarchical approach transforms the traditionally exponential “Navigation Among Movable Obstacles” (NAMO) problem into a computationally tractable task with linear complexity relative to the number of interactive objects. The efficacy of the algorithm is validated through extensive benchmarks in scenarios containing up to 45 interactive objects. Comparative results show that our method outperforms grid-based and sampling-based planners, achieving superior path quality and a higher execution success rate in high-fidelity simulations. Physical experiments using a perception-driven mobile platform further confirm the robustness and practical utility of the proposed method in unstructured real-world environments. Note to Practitioners—The autonomy of uncrewed systems in complex environments is limited by the static characteristics of the environment, which restricts their application scope and task execution capabilities. The method proposed in this paper enables mobile manipulator robotic systems to actively change the state of the environment through interaction, thereby enhancing their autonomous planning capabilities. To use this method, the geometric information of purely static obstacles and interactive elements in the scene, as well as the interaction models of these elements, such as pushing, pulling, and rotation, must be provided. Since this method explicitly plans the interaction process, it is necessary to model the kinematic characteristics of the agent and the interaction characteristics of the interactive elements. This method is suitable for multi-connected domain path planning problems that require consideration of scene interaction, such as uncrewed vehicle path planning tasks in disaster rescue, warehousing logistics, and other scenarios.