A Unified Experimental Architecture for Informative Path Planning: from Simulation to Deployment with GuadalPlanner

The evaluation of informative path planning algorithms for autonomous vehicles is often hindered by fragmented execution pipelines and limited transferability between simulation and real-world deployment. This paper introduces a unified architecture that decouples high-level decision-making from vehicle-specific control, enabling algorithms to be evaluated consistently across different abstraction levels without modification. The proposed architecture is realized through GuadalPlanner, which defines standardized interfaces between planning, sensing, and vehicle execution. It is an open and extensible research tool that supports discrete graph-based environments and interchangeable planning strategies, and is built upon widely adopted robotics technologies, including ROS2, MAVLink, and MQTT. Its design allows the same algorithmic logic to be deployed in fully simulated environments, software-in-the-loop configurations, and physical autonomous vehicles using an identical execution pipeline. The approach is validated through a set of experiments, including real-world deployment on an autonomous surface vehicle performing water quality monitoring with real-time sensor feedback.

💡 Research Summary

The paper addresses a persistent bottleneck in the evaluation of Informative Path Planning (IPP) algorithms: the lack of a unified execution pipeline that can seamlessly transition from simulation to real‑world deployment. The authors propose a modular architecture that cleanly separates high‑level decision‑making from low‑level vehicle control, enabling the same algorithmic code to run unchanged across three abstraction levels: (i) algorithm‑level simulation using idealized vehicle models in a graph‑based environment, (ii) software‑in‑the‑loop (SITL) where realistic control dynamics and communication delays are introduced via the ArduPilot SITL simulator, and (iii) real deployment on an autonomous surface vehicle (ASV) equipped with ArduPilot, ROS 2, and MQTT.

The implementation, named GuadalPlanner, defines standardized interfaces for planning, sensing, and fleet execution. A Graph Environment Monitor supplies a 2‑D graph representation, node uncertainties, and measurement models. The Planner Module implements any IPP strategy (the authors demonstrate greedy uncertainty reduction, Bayesian‑optimization‑based expected improvement, and submodular coverage) and interacts with the environment through a fixed API. The Fleet Interface bridges ROS 2 topics/services and MQTT messages to MAVLink commands, providing real‑time vehicle state (position, velocity, battery) to the planner and transmitting planned waypoints back to the vehicle. Backend adapters abstract the underlying platform: they can connect to a pure simulator, the ArduPilot SITL instance, or a physical vehicle without altering the planner code.

Experiments validate the architecture across all three levels. In simulation, the planners are evaluated on synthetic graphs to measure pure information gain and path efficiency. In SITL, the same code runs with realistic control loops, communication latency, and sensor noise, allowing assessment of algorithm robustness under near‑real conditions. Finally, the planners are deployed on a real ASV performing water‑quality monitoring; the vehicle streams sensor readings back to the planner, which dynamically replans to maximize information gain. Metrics such as total information collected, traveled distance, mission completion time, and energy consumption remain consistent across levels, demonstrating functional equivalence of the execution contexts.

A notable contribution is the support for multi‑agent scenarios via MQTT’s publish/subscribe model, enabling low‑power onboard computers to coordinate multiple ASVs for large‑scale monitoring tasks while preserving collision avoidance and global information sharing. The architecture is deliberately agnostic to vehicle type; only the backend adapter needs to be rewritten for UAVs, UGVs, or other platforms, making the framework extensible.

Compared with existing tools (Gazebo, AirSim, custom MATLAB/Python simulators), GuadalPlanner focuses on the high‑level IPP stack rather than low‑level dynamics alone, providing a reproducible, cross‑platform pipeline that bridges the simulation‑to‑reality gap. The open‑source release, ROS 2‑based modularity, and clear interface definitions facilitate community adoption and reproducibility.

In summary, the paper delivers a practical, well‑engineered solution for systematic, comparable, and transferable IPP experimentation. By decoupling planning from vehicle control and standardizing communication through ROS 2, MAVLink, and MQTT, GuadalPlanner enables researchers to iterate rapidly from algorithm design to field deployment without code rewrites, thereby accelerating the translation of IPP research into real‑world applications. Future work may extend the framework to three‑dimensional environments, richer sensor suites, and larger multi‑robot fleets.