Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Autonomous Navigation in Ice-Covered Waters with Learned Predictions on Ship-Ice Interactions

Ninghan Zhong, Alessandro Potenza, Stephen L. Smith

TL;DR

This work tackles autonomous navigation in ice-covered waters where collision-free paths are often infeasible due to dense, moving ice. It introduces a deep learning predictor that estimates coarse ice dynamics via occupancy maps, coupled with a physics-informed loss to enforce mass conservation, and a lattice-based planner that uses a caching-enabled A*-style search with an occupancy-difference collision cost. The approach is evaluated in both simulation and a physical testbed, showing reduced ice collisions and competitive travel distances compared to state-of-the-art baselines, especially at higher ice concentrations. The results demonstrate that anticipating obstacle motions can significantly enhance safety and efficiency for ASVs in icy environments, with public code and demos available for reproduction.

Abstract

Autonomous navigation in ice-covered waters poses significant challenges due to the frequent lack of viable collision-free trajectories. When complete obstacle avoidance is infeasible, it becomes imperative for the navigation strategy to minimize collisions. Additionally, the dynamic nature of ice, which moves in response to ship maneuvers, complicates the path planning process. To address these challenges, we propose a novel deep learning model to estimate the coarse dynamics of ice movements triggered by ship actions through occupancy estimation. To ensure real-time applicability, we propose a novel approach that caches intermediate prediction results and seamlessly integrates the predictive model into a graph search planner. We evaluate the proposed planner both in simulation and in a physical testbed against existing approaches and show that our planner significantly reduces collisions with ice when compared to the state-of-the-art. Codes and demos of this work are available at https://github.com/IvanIZ/predictive-asv-planner.

Autonomous Navigation in Ice-Covered Waters with Learned Predictions on Ship-Ice Interactions

TL;DR

This work tackles autonomous navigation in ice-covered waters where collision-free paths are often infeasible due to dense, moving ice. It introduces a deep learning predictor that estimates coarse ice dynamics via occupancy maps, coupled with a physics-informed loss to enforce mass conservation, and a lattice-based planner that uses a caching-enabled A*-style search with an occupancy-difference collision cost. The approach is evaluated in both simulation and a physical testbed, showing reduced ice collisions and competitive travel distances compared to state-of-the-art baselines, especially at higher ice concentrations. The results demonstrate that anticipating obstacle motions can significantly enhance safety and efficiency for ASVs in icy environments, with public code and demos available for reproduction.

Abstract

Autonomous navigation in ice-covered waters poses significant challenges due to the frequent lack of viable collision-free trajectories. When complete obstacle avoidance is infeasible, it becomes imperative for the navigation strategy to minimize collisions. Additionally, the dynamic nature of ice, which moves in response to ship maneuvers, complicates the path planning process. To address these challenges, we propose a novel deep learning model to estimate the coarse dynamics of ice movements triggered by ship actions through occupancy estimation. To ensure real-time applicability, we propose a novel approach that caches intermediate prediction results and seamlessly integrates the predictive model into a graph search planner. We evaluate the proposed planner both in simulation and in a physical testbed against existing approaches and show that our planner significantly reduces collisions with ice when compared to the state-of-the-art. Codes and demos of this work are available at https://github.com/IvanIZ/predictive-asv-planner.
Paper Structure (27 sections, 2 theorems, 15 equations, 5 figures, 1 table, 1 algorithm)

This paper contains 27 sections, 2 theorems, 15 equations, 5 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Contributions
  3. Related Work
  4. Problem Formulation
  5. Environment
  6. Ice Motion Prediction
  7. Path Planning
  8. Ice Floe Motion Prediction
  9. Space Representations
  10. Occupancy Prediction
  11. Model Training
  12. Training Data
  13. Loss Functions
  14. Path Planning Framework
  15. Lattice Planning with Motion Primitives
  16. ...and 12 more sections

Key Result

Theorem 1

Given a problem instance $I = (n_s, O_s, y_{\text{goal}})$, let $\textup{OPT}(I)$ denote the optimal path and let $\textup{Alg}(I)$ denote the path returned by Alg. alg:a_star_occ_diff assuming perfect occupancy predictions. Let $l_{\min}$ be the distance of the shortest action primitive, the cost o where $u$ and $\alpha$ are described in Eq. eq:path_cost_generic.

Figures (5)

  • Figure 1: The left panel shows a planned path from the proposed planner in simulation. The middle panels compare the predicted occupancy after path traversal with the ground truth occupancy. The right panel presents an experimental trial from the ship-ice navigation testbed at the Autonomous Systems Laboratory, University of Waterloo.
  • Figure 2: Ratio-based occupancy map is computed from the detected obstacles (a). Based on the ship's current pose (blue ship icon), a local occupancy observation (b) is cropped. Ship pose is encoded as footprint (c) and ship action is encoded as swath (d). U-Net is trained to predict the resulting occupancy. Predicted occupancy is stitched back into global occupancy.
  • Figure 3: A toy example showing occupancy differences. The motion primitives are simplified for illustration purposes. The red dotted boxes highlight the predicted motions of the obstacles. Here left swath is optimal.
  • Figure 4: Evaluation results from simulation across four concentrations
  • Figure 5: Testbed evaluations across 20%, 30%, and 40% concentrations.

Theorems & Definitions (3)

  • Theorem 1
  • Lemma 1
  • proof