Inspection Planning Primitives with Implicit Models

TL;DR

The paper tackles the memory bottleneck of sampling-based inspection planning for large, complex infrastructures by introducing IPIM, a framework that operates entirely on neural Signed Distance Functions. By replacing explicit environment models with a learned implicit representation and reworking the four core primitives—Collision Check, Observation Simulation, Observation Representation, and Total Coverage Check—IPIM dramatically reduces memory usage while preserving plan quality. Empirical results on scenarios including a real-world Plant with over 92 million mesh faces show memory savings up to 70x compared to state-of-the-art planners, with comparable coverage and planning performance. This approach enables scalable, memory-efficient inspection planning suitable for large-scale infrastructure monitoring and autonomous drone missions.

Abstract

The aging and increasing complexity of infrastructures make efficient inspection planning more critical in ensuring safety. Thanks to sampling-based motion planning, many inspection planners are fast. However, they often require huge memory. This is particularly true when the structure under inspection is large and complex, consisting of many struts and pillars of various geometry and sizes. Such structures can be represented efficiently using implicit models, such as neural Signed Distance Functions (SDFs). However, most primitive computations used in sampling-based inspection planner have been designed to work efficiently with explicit environment models, which in turn requires the planner to use explicit environment models or performs frequent transformations between implicit and explicit environment models during planning. This paper proposes a set of primitive computations, called Inspection Planning Primitives with Implicit Models (IPIM), that enable sampling-based inspection planners to entirely use neural SDFs representation during planning. Evaluation on three scenarios, including inspection of a complex real-world structure with over 92M triangular mesh faces, indicates that even a rudimentary sampling-based planner with IPIM can generate inspection trajectories of similar quality to those generated by the state-of-the-art planner, while using up to 70x less memory than the state-of-the-art inspection planner.

Figures (8)

• Figure 1: A glycol distillation plant at San Jacinto College, on which we evaluate the performance of IPIM. We name it Plant in Section \ref{['exp-set']}.
• Figure 2: Illustration of how IPIM is used with an inspection planner to reduce the memory cost. Before planning, the explicit environment $\mathcal{E}$ is converted to neural SDF $f_\mathcal{E}$ comprising a feature extractor (FE) and a Multi-Layer Perceptron (MLP). During planning, IPIM converts the four primitive computations of inspection planning with SDF to the implicit counterparts. Description about the four primitive computations is presented in Section \ref{['intro']}.
• Figure 3: From left to right: Bridge taken from fu2023asymptotically, and Plant-s taken as a subset of Plant shown in Fig. \ref{['fig: data']}.
• Figure 4: From left to right: planning time v.s. inspection coverage, and planning time v.s. cost of the inspection path. From up to bottom: scenario Bridge, scenario Plant-s and scenario Plant.
• Figure 5: Sample trajectories in Plant-s planned by TP-IPIM, with red covered, black uncovered and blue the inspection path. TP-IPIM plans within 2 hours (Up) and 3 hours (Bottom), with $55\%$ and $62\%$ coverage.