Field Testing of a Stochastic Planner for ASV Navigation Using Satellite Images

TL;DR

This work addresses robust long-term ASV navigation for lake water-quality monitoring by offline planning on satellite-derived stochastic graphs using PCCTPAO*, paired with a GPS-, vision-, and sonar-enabled local planner. The global planner computes a policy that minimizes expected travel distance under edge uncertainty, while the local stack disambiguates stochastic edges and safely tracks the policy via a fused occupancy grid and a lateral BIT*–MPC control loop. Field experiments on three km-scale missions demonstrate autonomous execution despite unmapped obstacles, with strong improvements over prior attempts and a clear set of lessons on traversal uncertainty, localization, and sensor fusion. The approach offers a practical pathway toward reliable, interpretable, and verifiable long-term autonomy for environmental monitoring with ASVs.

Abstract

We introduce a multi-sensor navigation system for autonomous surface vessels (ASV) intended for water-quality monitoring in freshwater lakes. Our mission planner uses satellite imagery as a prior map, formulating offline a mission-level policy for global navigation of the ASV and enabling autonomous online execution via local perception and local planning modules. A significant challenge is posed by the inconsistencies in traversability estimation between satellite images and real lakes, due to environmental effects such as wind, aquatic vegetation, shallow waters, and fluctuating water levels. Hence, we specifically modelled these traversability uncertainties as stochastic edges in a graph and optimized for a mission-level policy that minimizes the expected total travel distance. To execute the policy, we propose a modern local planner architecture that processes sensor inputs and plans paths to execute the high-level policy under uncertain traversability conditions. Our system was tested on three km-scale missions on a Northern Ontario lake, demonstrating that our GPS-, vision-, and sonar-enabled ASV system can effectively execute the mission-level policy and disambiguate the traversability of stochastic edges. Finally, we provide insights gained from practical field experience and offer several future directions to enhance the overall reliability of ASV navigation systems.

Figures (21)

• Figure 1: Real-world challenges that motivate the use of stochastic edges in our planning setup.
• Figure 2: A high-level overview of our navigation framework for water sampling. Given a set of user-selected target locations (red icons), our algorithm identifies stochastic edges from coarse satellite images and plans a mission-level policy for ASV navigation. Aerial views of two stochastic edges from real-world experiments are shown here.
• Figure 3: A toy example graph shown on the water mask generated from Sentinel-2 satellite images, with the corresponding graph on an aerial view image shown on the right. The planned paths between nodes are simplified for ease of understanding. The number beside each edge of the high-level graph is the path length in km, and the number in brackets is the blocking probability, which is computed using the probability of water coverage in each pixel (represented by its shade of orange) on the path. Note that traversable and ambiguous edges are the state before any action.
• Figure 4: The final AO tree after running PCCTP-AO* on the example in Fig. \ref{['fig:example_graph']}. The label inside each node is the current state of the robot. OR nodes are rectangles, and AND nodes are ellipses. Nodes that are part of the final policy are green, extra expanded nodes are yellow, and leaf nodes terminated early are orange. Some orange nodes that are terminated early are left out in this figure for simplicity. This figure is reproduced from Fig. 3 in boat2022.
• Figure 5: Example of how we relax the original problem graph to calculate the heuristic $h(n)$. At a high level, we construct a relaxed problem by removing all stochastic edges and unreachable nodes from the original graph. Then, the heuristic of the original problem is the cost of the relaxed problem and is always admissible.