pith. sign in

arxiv: 2604.08032 · v1 · submitted 2026-04-09 · 💻 cs.AI · cs.RO

"Why This Avoidance Maneuver?" Contrastive Explanations in Human-Supervised Maritime Autonomous Navigation

Joel Jose , Andreas Madsen , Andreas Brands{\ae}ter , Tor A. Johansen , Erlend M. Coates This is my paper

Pith reviewed 2026-05-10 17:50 UTC · model grok-4.3

classification 💻 cs.AI cs.RO
keywords contrastive explanationsmaritime autonomous navigationcollision avoidancehuman supervisionexplainable AIuser studycognitive workload
0
0 comments X

The pith

Contrastive explanations help marine officers grasp why an autonomous system picks one avoidance maneuver over others.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a framework that generates contrastive explanations for maritime collision avoidance by comparing the system's chosen maneuver against relevant alternatives, then presents these through visual and textual cues drawn from the underlying planner. An exploratory study with four experienced marine officers indicates that this approach improves supervisors' understanding of the system's objectives, especially in multi-vessel scenarios. The work also notes that the same explanations can raise cognitive workload, pointing toward selective, demand-driven delivery rather than always-on presentation. This matters because maritime autonomy will require human oversight for years, and supervisors need insight into complex causal logic to intervene effectively.

Core claim

The authors propose a method for generating contrastive explanations that compare a collision-avoidance system's proposed solution against relevant alternatives, rendered via visual and textual cues that highlight key objectives. In an exploratory user study, four experienced marine officers found these explanations supported understanding of the system's goals in complex encounters, yet the same cues increased cognitive workload, leading the authors to conclude that future interfaces may benefit most from demand-driven or scenario-specific explanation strategies.

What carries the argument

Contrastive explanation generation that pits the system's chosen maneuver against relevant alternatives, surfaced through visual and textual cues extracted from the state-of-the-art collision avoidance planner.

If this is right

  • Contrastive explanations prove especially useful in complex multi-vessel encounters.
  • The same explanations can increase cognitive workload for the supervisor.
  • Demand-driven or scenario-specific strategies for delivering explanations will likely outperform always-on presentation.
  • Human-supervised maritime autonomy requires selective transparency to balance insight and mental effort.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same contrastive approach could be adapted to other supervised autonomy domains where planners produce hard-to-intuit trajectories, such as drone traffic management.
  • Real-time workload monitoring could automatically trigger or suppress the explanations to keep supervisor attention within safe bounds.
  • Extending the method to include quantitative trade-off metrics (time saved versus risk incurred) might further reduce the observed workload increase.

Load-bearing premise

The visual and textual cues in the framework accurately and without bias highlight the key objectives from the underlying collision avoidance system.

What would settle it

A controlled follow-up study in which the same officers (or a larger group) perform navigation tasks with and without the contrastive cues and show no measurable gain in objective understanding or decision accuracy would falsify the central claim.

Figures

Figures reproduced from arXiv: 2604.08032 by Andreas Brands{\ae}ter, Andreas Madsen, Erlend M. Coates, Joel Jose, Tor A. Johansen.

Figure 1
Figure 1. Figure 1: Concept illustration. Contrastive explanations for a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The adapted PSW model from [24]. We believe that [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Contrastive Explanations Framework. The costs are [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The Visualization Interface used in this study. The design draws inspiration from the transparency layers in [6] and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Snapshots of traffic scenarios utilized in the user study, with the alternative maneuvers and associated explanations. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Responses from participants with management level [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
read the original abstract

Automated maritime collision avoidance will rely on human supervision for the foreseeable future. This necessitates transparency into how the system perceives a scenario and plans a maneuver. However, the causal logic behind avoidance maneuvers is often complex and difficult to convey to a navigator. This paper explores how to explain these factors in a selective, understandable manner for supervisors with a nautical background. We propose a method for generating contrastive explanations, which provide human-centric insights by comparing a system's proposed solution against relevant alternatives. To evaluate this, we developed a framework that uses visual and textual cues to highlight key objectives from a state-of-the-art collision avoidance system. An exploratory user study with four experienced marine officers suggests that contrastive explanations support the understanding of the system's objectives. However, our findings also reveal that while these explanations are highly valuable in complex multi-vessel encounters, they can increase cognitive workload, suggesting that future maritime interfaces may benefit most from demand-driven or scenario-specific explanation strategies.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a method for generating contrastive explanations that compare a maritime collision avoidance system's proposed maneuver against relevant alternatives, aiming to provide human-centric insights into complex decision factors. It develops a framework using visual and textual cues to highlight key objectives from a state-of-the-art avoidance system and evaluates this via an exploratory user study with four experienced marine officers. The study suggests that such explanations support understanding of the system's objectives, especially in multi-vessel encounters, while noting potential increases in cognitive workload and recommending demand-driven explanation strategies for future interfaces.

Significance. If the contrastive explanations are shown to faithfully capture the underlying system's logic and the preliminary findings generalize beyond the small sample, this could advance human-AI collaboration in safety-critical maritime navigation by making avoidance maneuvers more interpretable to supervisors with nautical expertise. The work provides domain-specific application of contrastive explanations and initial expert feedback, which strengthens its relevance for applied AI in transportation.

major comments (2)
  1. [Abstract] Abstract: The central claim that contrastive explanations 'support the understanding of the system's objectives' depends on the assumption that the framework's visual and textual cues accurately and without bias surface the true decision factors from the collision avoidance algorithm. The exploratory study with four participants captures only subjective impressions of helpfulness and provides no objective evidence (e.g., via sensitivity analysis, cost-term comparison, or constraint-priority matching) that the highlighted elements align with the system's internal logic rather than a curated subset. This is load-bearing for the claim, as biased cues could produce illusory understanding.
  2. [Abstract] Abstract and evaluation: With only four participants and no reported details on study design, controls, tasks, statistical analysis, or exact explanation generation method, the findings remain preliminary. This limits the ability to substantiate benefits in complex scenarios or workload increases, requiring either expanded validation or stronger qualification of the results.
minor comments (1)
  1. [Abstract] Abstract: Lacks specifics on participant selection criteria, exact scenarios tested, quantitative measures for understanding and workload, and how the contrastive explanations were generated from the underlying system.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We agree that the exploratory nature of the study and the reliance on subjective impressions warrant stronger qualification of the claims in the abstract and throughout the paper. We will revise the manuscript to address the concerns about objective evidence for alignment with the system's logic and to better contextualize the preliminary findings.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that contrastive explanations 'support the understanding of the system's objectives' depends on the assumption that the framework's visual and textual cues accurately and without bias surface the true decision factors from the collision avoidance algorithm. The exploratory study with four participants captures only subjective impressions of helpfulness and provides no objective evidence (e.g., via sensitivity analysis, cost-term comparison, or constraint-priority matching) that the highlighted elements align with the system's internal logic rather than a curated subset. This is load-bearing for the claim, as biased cues could produce illusory understanding.

    Authors: We acknowledge the validity of this concern. The current evaluation relies on subjective expert impressions rather than objective measures of fidelity to the underlying algorithm. The contrastive explanations are derived directly from the system's cost function and constraint priorities (as described in Section 3), but we did not include sensitivity analysis or explicit cost-term matching in the study. We will revise the abstract to qualify the central claim as 'suggest that contrastive explanations are perceived to support understanding of the system's objectives based on expert feedback' and expand the methods and discussion sections to explicitly link the visual/textual cues to specific cost terms and constraints. We will also note the absence of objective validation as a limitation and a direction for future work. revision: partial

  2. Referee: [Abstract] Abstract and evaluation: With only four participants and no reported details on study design, controls, tasks, statistical analysis, or exact explanation generation method, the findings remain preliminary. This limits the ability to substantiate benefits in complex scenarios or workload increases, requiring either expanded validation or stronger qualification of the results.

    Authors: We agree that the study is preliminary and that the abstract and evaluation section should be strengthened with additional details and qualifications. The manuscript already labels the work as exploratory, but we will add explicit descriptions of the study protocol, tasks (scenario walkthroughs with think-aloud protocol), qualitative analysis approach, and the exact method for generating explanations (comparison of cost terms across maneuvers). No statistical analysis was conducted due to the small sample, which we will state clearly. We will revise the abstract and results to emphasize that the findings are suggestive rather than conclusive, particularly regarding workload increases in multi-vessel scenarios, and recommend demand-driven strategies as a cautious interpretation. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical user study with no derivations or self-referential predictions

full rationale

The paper describes development of a contrastive explanation framework for a maritime collision avoidance system followed by an exploratory user study with four marine officers. No mathematical derivations, equations, fitted parameters, or predictions are present in the provided text or abstract. The central claim rests on subjective participant feedback rather than any reduction of outputs to inputs by construction. Self-citations, if present, are not load-bearing for the empirical findings. This is a standard non-circular application study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is an applied empirical study in human-AI interaction with no mathematical model, free parameters, or new postulated entities.

pith-pipeline@v0.9.0 · 5477 in / 1043 out tokens · 38277 ms · 2026-05-10T17:50:14.159666+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    Operational and economic advantages of autonomous ships and their perceived impacts on port operations

    Ismail Kurt and Murat Aymelek. Operational and economic advantages of autonomous ships and their perceived impacts on port operations. Maritime Economics & Logistics, 24, 2022

    work page 2022

  2. [2]

    Path planning and collision avoidance for autonomous surface vehicles ii: a comparative study of algorithms.Journal of Marine Science and Technology, 2021

    Anete Vagale, Robin Bye, Rachid Oucheikh, Ottar Osen, and Thor Fossen. Path planning and collision avoidance for autonomous surface vehicles ii: a comparative study of algorithms.Journal of Marine Science and Technology, 2021

    work page 2021

  3. [3]

    Towards approval of autonomous ship systems by their operational envelope.Journal of Marine Science and Technology, 27(1):67–76, 2022

    Ørnulf Jan Rødseth, Lars Andreas Lien Wennersberg, and H ˚avard Nordahl. Towards approval of autonomous ship systems by their operational envelope.Journal of Marine Science and Technology, 27(1):67–76, 2022

    work page 2022

  4. [4]

    Improving decision transparency in autonomous maritime collision avoidance.Journal of Marine Science and Tech- nology, 2025

    Andreas Nygard Madsen, Andreas Brandsæter, Koen van de Merwe, and Jooyoung Park. Improving decision transparency in autonomous maritime collision avoidance.Journal of Marine Science and Tech- nology, 2025

    work page 2025

  5. [5]

    Mica R. Endsley. Supporting Human-AI Teams:Transparency, ex- plainability, and situation awareness.Computers in Human Behavior, 140:107574, 2023

    work page 2023

  6. [6]

    The Influence of Agent Transparency and Complexity on Situation Awareness, Mental Workload, and Task Performance

    Koen Van De Merwe, Steven Mallam, Salman Nazir, and Øystein Engelhardtsen. The Influence of Agent Transparency and Complexity on Situation Awareness, Mental Workload, and Task Performance. Journal of Cognitive Engineering and Decision Making, 18(2):156– 184, 2024

    work page 2024

  7. [7]

    Toward a Theory of Situation Awareness in Dynamic Systems.Human Factors: The Journal of the Human Factors and Ergonomics Society, 37:32–64, 1995

    Mica Endsley. Toward a Theory of Situation Awareness in Dynamic Systems.Human Factors: The Journal of the Human Factors and Ergonomics Society, 37:32–64, 1995

    work page 1995

  8. [8]

    A model for types and levels of human interaction with automation

    Raja Parasuraman, Thomas Sheridan, and Christopher Wickens. A model for types and levels of human interaction with automation. ieee trans. syst. man cybern. part a syst. hum. 30(3), 286-297.IEEE transactions on systems, man, and cybernetics. Part A, Systems and humans : a publication of the IEEE Systems, Man, and Cybernetics Society, 30:286–97, 06 2000

    work page 2000

  9. [9]

    A simulator-based ap- proach for testing and assessing human supervised autonomous ship navigation.Journal of Marine Science and Technology, 29(2):432– 445, 2024

    Andreas Brandsæter and Andreas Madsen. A simulator-based ap- proach for testing and assessing human supervised autonomous ship navigation.Journal of Marine Science and Technology, 29(2):432– 445, 2024

    work page 2024

  10. [10]

    Explanations in Autonomous Driving: A Survey.IEEE Transactions on Intelligent Transportation Systems, 23(8):10142–10162, 2022

    Daniel Omeiza, Helena Webb, Marina Jirotka, and Lars Kunze. Explanations in Autonomous Driving: A Survey.IEEE Transactions on Intelligent Transportation Systems, 23(8):10142–10162, 2022

    work page 2022

  11. [11]

    Gjærum, Inga Str ¨umke, Anastasios M

    Vilde B. Gjærum, Inga Str ¨umke, Anastasios M. Lekkas, and Timothy Miller. Real-Time Counterfactual Explanations For Robotic Systems With Multiple Continuous Outputs.IFAC-PapersOnLine, 56(2):7–12,

    work page

  12. [12]

    Publisher: Elsevier BV

    work page

  13. [13]

    Explaining a Deep Reinforcement Learning Docking Agent Using Linear Model Trees with User Adapted Visualization.Marine Science and Engineering, 2021

    Vilde B Gjaerum, Inga Str ¨umke, Ole Andreas Alsos, and Anastasios M Lekkas. Explaining a Deep Reinforcement Learning Docking Agent Using Linear Model Trees with User Adapted Visualization.Marine Science and Engineering, 2021

    work page 2021

  14. [14]

    Osen, and Robin T

    Vijander Singh, Ottar L. Osen, and Robin T. Bye. Explainable artificial intelligence for autonomous surface vessels by fuzzy-based collision avoidance system. In Tomonobu Senjyu, Chakchai So-In, and Amit Joshi, editors,Smart Trends in Computing and Communications, Singapore, 2023. Springer Nature Singapore

    work page 2023

  15. [15]

    Human-centered explainable artificial intelligence for marine autonomous surface vehicles.Journal of Marine Science and Engineering, 9(11), 2021

    Erik Veitch and Ole Andreas Alsos. Human-centered explainable artificial intelligence for marine autonomous surface vehicles.Journal of Marine Science and Engineering, 9(11), 2021

    work page 2021

  16. [16]

    Contrastive explanation.Royal Institute of Philosophy Supplement, 27:247–266, 1990

    Peter Lipton. Contrastive explanation.Royal Institute of Philosophy Supplement, 27:247–266, 1990

    work page 1990

  17. [17]

    Explanation in artificial intelligence: Insights from the social sciences.Artificial Intelligence, 267:1–38, 2019

    Tim Miller. Explanation in artificial intelligence: Insights from the social sciences.Artificial Intelligence, 267:1–38, 2019

    work page 2019

  18. [18]

    Huang, and X

    Ruikun Luo, Na Du, Kevin Y . Huang, and X. Jessie Yang. Enhanc- ing transparency in human-autonomy teaming via the option-centric rationale display.Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 63(1):166–167, 2019

    work page 2019

  19. [19]

    Rohlfing, and Britta Wrede

    Andr ´e Groß, Amit Singh, Ngoc Chi Banh, Birte Richter, Ingrid Scharlau, Katharina J. Rohlfing, and Britta Wrede. Scaffolding the human partner by contrastive guidance in an explanatory human-robot dialogue.Frontiers in Robotics and AI, 10, 2023. Publisher: Frontiers

    work page 2023

  20. [20]

    Benjamin Krarup, Senka Krivic, Daniele Magazzeni, Derek Long, Michael Cashmore, and David E. Smith. Contrastive Explanations of Plans through Model Restrictions.Journal of Artificial Intelligence Research, 72:533–612, 2021

    work page 2021

  21. [21]

    Towards Accountability: Providing Intelligible Explanations in Autonomous Driving

    Daniel Omeiza, Helena Web, Marina Jirotka, and Lars Kunze. Towards Accountability: Providing Intelligible Explanations in Autonomous Driving. In2021 IEEE Intelligent Vehicles Symposium (IV), 2021

    work page 2021

  22. [22]

    Towards providing explanations for robot motion planning

    Martim Brand ˜ao, Gerard Canal, Senka Krivi´c, and Daniele Magazzeni. Towards providing explanations for robot motion planning. In2021 IEEE International Conference on Robotics and Automation (ICRA),

    work page

  23. [23]

    CE-MRS: Contrastive Explanations for Multi-Robot Systems

    Ethan Schneider, Daniel Wu, Devleena Das, and Sonia Cher- nova. CE-MRS: Contrastive Explanations for Multi-Robot Systems. IEEE Robotics and Automation Letters, 9(11):10121–10128, 2024. arXiv:2410.08408 [cs]

    work page arXiv 2024

  24. [24]

    Collaborative collision avoidance for maritime autonomous surface ships: A review

    Melih Akda ˘g, Petter Solnør, and Tor Arne Johansen. Collaborative collision avoidance for maritime autonomous surface ships: A review. Ocean Engineering, 2022

    work page 2022

  25. [25]

    Koen van de Merwe, Steven Mallam, Øystein Engelhardtsen, and Salman Nazir. Towards an approach to define transparency require- ments for maritime collision avoidance.Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 67(1):483–488, September 2023

    work page 2023

  26. [26]

    Tor Arne Johansen, Tristan Perez, and Andrea Cristofaro. Ship Collision Avoidance and COLREGS Compliance Using Simulation- Based Control Behavior Selection With Predictive Hazard Assess- ment.IEEE Transactions on Intelligent Transportation Systems, 17(12):3407–3422, 2016

    work page 2016

  27. [27]

    Hagen, D.K.M

    I.B. Hagen, D.K.M. Kufoalor, T.A. Johansen, and E.F. Brekke. Scenario-Based Model Predictive Control with Several Steps for COLREGS Compliant Ship Collision Avoidance.IFAC-PapersOnLine, 55(31):307–312, 2022

    work page 2022

  28. [28]

    Explainability in deep reinforcement learning.Knowledge-Based Systems, 214, 2021

    Alexandre Heuillet, Fabien Couthouis, and Natalia D ´ıaz-Rodr´ıguez. Explainability in deep reinforcement learning.Knowledge-Based Systems, 214, 2021

    work page 2021

  29. [29]

    Sk ˚aden, Ole Andreas Alsos, Andreas Madsen, and Thomas Porathe

    Philip Hodne, Oskar K. Sk ˚aden, Ole Andreas Alsos, Andreas Madsen, and Thomas Porathe. Conversational user interfaces for maritime autonomous surface ships.Ocean Engineering, 310:118641, 2024

    work page 2024

  30. [30]

    From captain to button-presser: operators’ perspectives on navigating highly automated ferries.Journal of Physics: Conference Series, 2311:012028, 07 2022

    Erik Veitch, Kim Christensen, Markus Log, Erik Valestrand, Sigurd Hilmo Lundheim, Martin Nesse, Ole Alsos, and Martin Steinert. From captain to button-presser: operators’ perspectives on navigating highly automated ferries.Journal of Physics: Conference Series, 2311:012028, 07 2022

    work page 2022

  31. [31]

    Redesigning ui for teleoperating rov using apple vision pro: A study on usability and workload

    Henrik Viken Lied, Taufik Akbar Sitompul, and Ole Andreas Alsos. Redesigning ui for teleoperating rov using apple vision pro: A study on usability and workload. InProceedings of the 20th International Conference on Virtual Reality Continuum and Its Applications in Industry. ACM, December 2025

    work page 2025

  32. [32]

    Johansen

    Trym Tengesdal and Tor A. Johansen. Simulation Framework and Software Environment for Evaluating Automatic Ship Collision Avoid- ance Algorithms. In2023 IEEE Conference on Control Technology and Applications (CCTA), Bridgetown, Barbados, 2023. IEEE

    work page 2023

  33. [33]

    I. B. Hagen, D. K. M. Kufoalor, E. F. Brekke, and T. A. Johansen. Mpc-based collision avoidance strategy for existing marine vessel guidance systems. In2018 IEEE International Conference on Robotics and Automation (ICRA), 2018

    work page 2018

  34. [34]

    Morten Breivik and Thor I. Fossen. Guidance laws for planar motion control. In2008 47th IEEE Conference on Decision and Control, pages 570–577, 2008

    work page 2008

  35. [35]

    Dirk M. Elston. The novelty effect.Journal of the American Academy of Dermatology, 85:565–566, 2021

    work page 2021

  36. [36]

    Understanding automation transparency and its adaptive design implications in safety–critical systems.Safety Science, 184:106730, 2025

    Mina Saghafian, Dorthea Mathilde Kristin Vatn, Stine Thordarson Moltubakk, Lene Elisabeth Bertheussen, Felix Marcel Petermann, Stig Ole Johnsen, and Ole Andreas Alsos. Understanding automation transparency and its adaptive design implications in safety–critical systems.Safety Science, 184:106730, 2025

    work page 2025