OmniPlanner: Universal Exploration and Inspection Path Planning across Robot Morphologies

TL;DR

OmniPlanner is presented, a unified planning framework for autonomous exploration and inspection across aerial, ground, and underwater robots that integrates volumetric exploration and viewpoint-based inspection, alongside target reach behaviors within a single modular architecture.

Abstract

Autonomous robotic systems are increasingly deployed for mapping, monitoring, and inspection in complex and unstructured environments. However, most existing path planning approaches remain domain-specific (i.e., either on air, land, or sea), limiting their scalability and cross-platform applicability. This article presents OmniPlanner, a unified planning framework for autonomous exploration and inspection across aerial, ground, and underwater robots. The method integrates volumetric exploration and viewpoint-based inspection, alongside target reach behaviors within a single modular architecture, complemented by a platform abstraction layer that captures morphology-specific sensing, traversability and motion constraints. This enables the same planning strategy to generalize across distinct mobility domains with minimal retuning. The framework is validated through extensive simulation studies and field deployments in underground mines, industrial facilities, forests, submarine bunkers, and structured outdoor environments. Across these diverse scenarios, OmniPlanner demonstrates robust performance, consistent cross-domain generalization, and improved exploration and inspection efficiency compared to representative state-of-the-art baselines.

Figures (20)

• Figure 1: An overview of the omniplanner core functionalities and features along with instances of field deployments. omniplanner has been deployed across aerial, ground, and underwater robots in a diverse set of environments. Various aspects of the planner, such as the planning behaviors, embodiment-specific adaptations, and the bifurcated local-global is shown through instances of field-deployments.
• Figure 2: omniplanner -- a platform-agnostic planning kernel supports global and local planning, while adaptation layers abstract robot embodiments and map representations. Task-specific behaviors are instantiated as objectives and features on top of the shared kernel, enabling reusable autonomy across heterogeneous platforms.
• Figure 3: Indicative visualization of local sampling strategies within the bounded planning volume. Green points denote collision-free valid samples, yellow points indicate invalid samples rejected during collision checking, and the underlying point cloud represents the current volumetric map. (a) Uniform sampling distributes configurations evenly throughout the local bounding box, promoting broad spatial coverage. (b) Gaussian sampling concentrates samples around the current robot configuration, yielding dense local connectivity in constrained regions. (c) Hybrid sampling combines Uniform and Gaussian distributions to balance local maneuverability and global reach.
• Figure 4: Comparison of local graph construction strategies in a multi-room building environment. Top-left: average number of rooms covered as a function of computation time for Basic and Batch graph construction. For each strategy, experiments were performed with eight different numbers of sampled points, each repeated over 20 trials. The plotted curves report the mean computation time and the mean number of reachable rooms covered by the local graph. Top-right: layout and dimensions of the evaluation environment. Bottom: representative local graph instances generated by the two strategies within the same planning volume.
• Figure 5: Comparison of local graph construction strategies in a multi-branch mine environment. Top-left: average number of branches covered as a function of computation time for Basic and Batch graph construction. For each strategy, experiments were performed with five different numbers of sampled points, each repeated over 20 trials. The plotted curves report the mean computation time and the mean number of reachable branches covered by the local graph. Top-right: layout and dimensions of the evaluation environment. Bottom: representative local graph instances generated by the two strategies within the same planning volume.

Theorems & Definitions (2)

• Definition 1: Residual Volume
• Definition 2: Residual Surface