Agnes Ngina Mwange君学位論文公聴会スライド　On the Empirical Inference of Ship Maneuvering Characteristics for Autonomous Berthing and Unberthing

DOCTORAL DISSERTATION On the Empirical Inference of Ship Maneuvering Characteristics for Autonomous Berthing and Unberthing Trajectory Planning, Low-Speed Maneuvering Modeling and Adaptive Ship Domain July 2026 Department of Naval Architecture & Ocean Engineering Division of Global Architecture Graduate School of Engineering The University of Osaka Agnes Ngina Mwange Supervisor: Professor Atsuo Maki

OUTLINE PART 00 CONTRIBUTION 1 CONTRIBUTION 2 CONTRIBUTION 3 PART 04 00 01 02 03 04 Introduction Trajectory planner Low-speed model Adaptive ship domain Concluding remarks o The berthing and/or o Overview o Overview o Overview o Concluding remarks o OCP → NLP → SQP o Model derivation o Spatial o Limitations o Deceleration o System unberthing problem o Degrees of autonomy o The safety stakes o Research scope guidelines o Collision avoidance o Feasibility and robustness identification o Model validation maneuvering o Future work conditions o Publications o Model definition o Model validation o Empirical vs analytical domain 02

PART 00 Introduction 03

BERTHING AND UNBERTHING UNBERTHING — DEPARTURE BERTHING — ARRIVAL • Removal of mooring ropes • Departure pose • Unberthing approach phase: o acceleration, o reduced steerability, o reduced disturbances counterattacking capacity o highly nonlinear dynamics, o constrained maneuvering space • Port exit • Port entry • Berthing approach phase: o deceleration, o reduced steerability, o reduced disturbances counterattacking capacity o highly nonlinear dynamics, o constrained maneuvering space • Docking pose • Alignment phase: o crabbing o mooring 04

MARINE INCIDENTS AND ACCIDENTS Key challenges associated with traditional berthing and unberthing operations: ~50% ~90% ~20% of marine accidents & incidents are of marine incidents involve some of maritime accidents are collisions[2]. attributed to human error [1]. form of loss of control[2] . 1. European Maritime Safety Agency (EMSA). “Annual Overview of Marine Casualties and Incidents2023.” In: (2023). url: https://emsa.europa.eu/publications/download/7639/5052/23.html (visited on 12/08/2024) 2. Japan Transport Safety Board (JTSB). JTSB Annual Report 2021. 2021. url: https://www.mlit.go.jp/jtsb/jtsbannualreport2021.html 05

• https://emsa.europa.eu/publications/download/7639/5052/23.html
• https://www.mlit.go.jp/jtsb/jtsbannualreport2021.html

AUTONOMOUS BERTHING AND UNBERTHING Degrees of autonomy[3] DEGREE 1 DEGREE 2 DEGREE 3 DEGREE 4 Decision support Remote, crewed Remote, uncrewed Fully autonomous • Automated processes • Controlled from ashore, • No crew on board • The operating system • Seafarers on board with crew aboard as a • An operations center operating and controlling systems. fallback. controls the ship remotely. decides and acts entirely by itself. • This is the focus of this research. 3. International Maritime Organization (IMO). Outcome of the regulatory scoping exercise for the use of Maritime Autonomous Surface Ships (MASS). 2021. url: https://wwwcdn.imo.org/localresources/en/MediaCentre/PressBriefings/Documents/MSC.1-Circ.1638%20%20Outcome%20Of%20The%20Regulatory%20Scoping%20ExerciseFor%20The%20Use%20Of%20Maritime%20Autonomous%20Surface%20Ships...%20(Secretariat).pdf 06

• https://wwwcdn.imo.org/localresources/en/MediaCentre/PressBriefings/Documents/MSC.1-Circ.1638 - Outcome Of The Regulatory Scoping ExerciseFor The Use Of Maritime Autonomous Surface Ships... (Secretariat).pdf

AUTONOMOUS BERTHING AND UNBERTHING PROBLEM 01 03 Low-speed dynamics 02 Confined water Rudder effectiveness collapses with surge velocity; Vertical limits from shallow water (under-keel the ship becomes underactuated and needs thrusters clearance) and horizontal limits from quays, buoys to control surge, sway and yaw independently. and traffic must be respected simultaneously. Disturbances dominate 04 Safety is regulated At near-zero speed, wind and current forces can rival COLREGs Rule 6 mandates a safe speed at all or exceed the ship's own control capacity. times[4], preserving the ability to execute emergency maneuvers. Ports worldwide also implement reduced speed zones (RSZs). 4. International Maritime Organization (IMO). COLREG: Convention on the International Regulations for Preventing Collisions at Sea, 1972. IMO Publication. International Maritime Organization, 2003. isbn:9789280141672. url: https://books.google.co.jp/books?id=%5C_ZkZAQAAIAAJ 07

• https://books.google.co.jp/books?id=\_ZkZAQAAIAAJ

AUTONOMOUS BERTHING AND UNBERTHING RESEARCH AREAS Trajectory planning Perception and sensing Focus: Path generation from approach fairway to berth position. Key Challenges: Obstacle avoidance, realtime replanning, curved trajectories Motion control Focus: Low-speed maneuverability and tracking control. Key Challenges: Actuator saturation, counteracting environmental disturbances, propeller reversal Maneuvering models Focus: Low-speed maneuvering prediction Key Challenges: Shallow water/bank effects, propeller-hull-rudder interaction Focus: Ship-shore relative positioning and environmental detection Key Challenges: GPS-denied harbors, allweather reliability, multi-sensor fusion KEY RESEARCH AREAS Human-ship interaction Focus: Remote monitoring and takeover systems Key Challenges: Trust calibration, situation awareness, minimal crew contexts Verification and validation Focus: Simulation-to-full-scale transfer Key Challenges: Standardized scenarios, performance metrics, data acquisition 08

RESEARCH GAPS Trajectory planning Maneuvering models • Only about 1% of path/trajectory-planning algorithms address berthing [5] • Berthing trajectory planning does not address deceleration profiles • Low-speed models are often identified from data far narrower than berthing conditions [6,7] • Low-speed models based on model tests introduce model-to-full-scale translation errors • Complex models limit the real-time implementation of trajectory planners Collision avoidance • Most safety envelop models (ship domains) focus on open sea or channel navigation • Analytical ship domains lower collision probability but overlook ship-specific and port-specific maneuvering behavior. Verification and validation • Model-to-full-scale translation uncertainties associated with model-based models and algorithms 5. Ülkü Öztürk, Melih Akdağ, and Tarık Ayabakan. “A review of path planning algorithms in maritime autonomous surface ships: Navigation safety perspective.” In: Ocean Engineering 251 (2022), p. 111010.issn: 0029-8018. doi: https://doi.org/10.1016/j.oceaneng.2022.111010 6. Yoshiki Miyauchi, Atsuo Maki, Naoya Umeda, Dimas M. Rachman, and Youhei Akimoto. “System parameter exploration of ship maneuvering model for automatic docking/berthing using CMA-ES.” In: Journal of Marine Science and Technology (June 2022). issn: 0948-4280. doi: 10.1007/s00773-022-00889-3. url: http://arxiv.org/abs/2111.06124%20https://link.springer.com/article/10.1007/s00773-022-008893%20https://link.springer.com/10.1007/s00773-022-00889-3 7. Agnes N. Mwange, Yoshiki Miyauchi, Taichi Kambara, Hiroaki Koike, Kazuyoshi Hosogaya, Atsushi Ishibashi, and Atsuo Maki. “Quantitative evaluation of full-scale ship maneuvering characteristics during berthing and unberthing.” In: Journal of Marine Science and Technology (2025). issn: 09484280. doi:10.1007/s00773-025-01098-4 09

• https://doi.org/10.1016/j.oceaneng.2022.111010
• http://arxiv.org/abs/2111.06124 https:/link.springer.com/article/10.1007/s00773-022-00889-3 https:/link.springer.com/10.1007/s00773-022-00889-3

HYPOTHESIS: Key challenges in autonomous berthing and unberthing can be addressed by systematically understanding manual navigation , then replicating and/or enhancing those practices in autonomous operations. 10

PART 01 · CONTRIBUTION 1 Trajectory Planner 11

OVERVIEW TRAJECTORY PLANNER Berthing approach phase: 1. Ship dynamics 2. Actuation capacity 3. Environmental disturbances 4. Ship deceleration 5. Collision avoidance 6. Berthing time Successful trajectory planning: generation of dynamically and practically feasible, safe trajectories that meet predetermined objectives. Difference between path planning and trajectory planning: Path planning: geometric determination of a spatial route from start to goal without considering time or dynamics (kinematic feasibility) Trajectory planning: extends path planning by incorporating temporal information such as velocities and dynamic constraints along the path. 12

SUBJECT SHIP – PRINCIPAL PARTICULARS Ship A TRAJECTORY PLANNER Ship B 13

SUBJECT SHIP – COORDINATE SYSTEMS Ship A TRAJECTORY PLANNER Ship B Relationship between the two coordinate systems: 14

SUBJECT SHIP – EQUATIONS OF MOTION (Factor 1) TRAJECTORY PLANNER 8. Yasuo Yoshimura, Ikao Nakao, and Atsushi Ishibashi. “Unified Mathematical Model for Ocean and Harbour Manoeuvring.” In: International Conference on Marine Simulation and Ship Maneuverability, Aug. 2009, pp. 116–124. url: http://hdl.handle.net/2115/42969 9. H. Yasukawa and Y. Yoshimura. “Introduction of MMG standard method for ship maneuvering pre-dictions.” In: Journal of Marine Science and Technology (Japan) 20 (1 Mar. 2015), pp. 37–52. issn:09484280. doi: 10.1007/s00773-014-0293-y 10. Donghoon Kang, Vishwanath Nagarajan, Kazuhiko Hasegawa, and Masaaki Sano. “Mathematical model of single-propeller twin-rudder ship.” In: Journal of marine science and technology 13 (2008), pp. 207– 222. 11. Toshifumi Fujiwara, Michio Ueno, and Tadashi Nimura. “Estimation of wind forces and moments acting on ships.” In: Journal of the Society of Naval Architects of Japan 1998.183 (1998), pp. 77–90. 15

• http://hdl.handle.net/2115/42969

TRAJECTORY OPTIMIZATION TRAJECTORY PLANNER Trajectory optimization Analytical / indirect Direct numerical Stochastic / heuristic Optimality conditions Discretize, then solve NLP Search, no gradients needed Pontryagin's maximum principle Direct collocation Genetic algorithms Calculus of variations Direct shooting Particle swarm Dynamic programming Pseudospectral methods Ant colony / A* search Isochrone method Model predictive control Simulated annealing 16

TRAJECTORY OPTIMIZATION – OPTIMAL CONTROL PROBLEM TRAJECTORY PLANNER where, 12. Atsuo Maki, Youhei Akimoto, and Umeda Naoya. “Application of optimal control theory based on the evolution strategy (CMA-ES) to automatic berthing (part: 2).” In: Journal of Marine Science and Technology 26 (2021), pp. 835–845 13. Dimas M Rachman, Atsuo Maki, Yoshiki Miyauchi, and Naoya Umeda. “Warm-started semi online trajectory planner for ship’ s automatic docking (berthing).” In: Ocean Engineering 252 (2022), p. 111127 17

OCP – DIRECT MULTIPLE SHOOTING TRAJECTORY PLANNER • Time grid 𝑡0 , 𝑡f is discretized into 𝑁 segments: 𝑡0 , 𝑡1 , … , 𝑡𝑁−1 , 𝑡f . Let 𝑖 denote the 𝑖th segment such that, 𝑖 = 1,2, … 𝑁. • Let 𝑘 denote the end point of the segments, also denoted as knots, such that 𝑘 = 1,2, … . 𝑁𝑘 . 𝑁𝑘 = 𝑁+1 • Discretize states and control variables • Integrate along each segment and introduce constraints to reduce defects between segments • For any consecutive segments, constraints require: 𝐱𝑘− = 𝐱𝑘+ • Pro: breaks sensitivity present in single shooting 18

ACTUATOR LIMITS (Factor 2 and 3) TRAJECTORY PLANNER Actuator (control limits) (𝑢min , 𝑢max ): (i) Port rudder angles: δp,min = −1050 , δp,max = −450 (ii) Starboard rudder angles: δs,min = 450 , δs,max = 1050 (iii) Propeller revolution: 𝑛p,min = 0, 𝑛p,max = 10rps (iv) Bow-thruster revolution: 𝑛bt,min = −30rps, 𝑛bt,max = 30rps 7. Agnes N. Mwange, Yoshiki Miyauchi, Taichi Kambara, Hiroaki Koike, Kazuyoshi Hosogaya, Atsushi Ishibashi, and Atsuo Maki. “Quantitative evaluation of full-scale ship maneuvering characteristics during berthing and unberthing.” In: Journal of Marine Science and Technology (2025). issn: 09484280. doi:10.1007/s00773-025-01098-4 19

SHIP DECELERATION (Factor 4) TRAJECTORY PLANNER Maritime ports worldwide implement reduced speed zones (RSZs) where ship should maintain a set speed or less. ~10 knots 14. K Inoue, SETA Hiroaki, and MASUDA Kenji. “Guidelines for Speed Reduction in Berthing Manoeuvre.” In: The Journal of Japan Institute of Navigation (2002), pp. 169–176. 20

SHIP DECELERATION TRAJECTORY PLANNER Ship deceleration speed limits equation: 12. Atsuo Maki, Youhei Akimoto, and Umeda Naoya. “Application of optimal control theory based on the evolution strategy (CMA-ES) to automatic berthing (part: 2).” In: Journal of Marine Science and Technology 26 (2021), pp. 835–845 13. Dimas M Rachman, Atsuo Maki, Yoshiki Miyauchi, and Naoya Umeda. “Warm-started semi online trajectory planner for ship’ s automatic docking (berthing).” In: Ocean Engineering 252 (2022), p. 111127 21

COLLISION AVOIDANCE (Factor 5) TRAJECTORY PLANNER Point in polygon problem 13. Dimas M Rachman, Atsuo Maki, Yoshiki Miyauchi, and Naoya Umeda. “Warm-started semionline trajectory planner for ship’ s automatic docking (berthing).” In: Ocean Engineering 252 (2022), p. 111127 15. Yoshiki Miyauchi, Ryohei Sawada, Youhei Akimoto, Naoya Umeda, and Atsuo Maki. “Optimization on planning of trajectory and control of autonomous berthing and unberthing for the realistic port geometry.” In: Ocean Engineering 245 (2022), p. 110390. issn: 0029-8018. doi: https://doi.org/10.1016/j.oceaneng.2021.110390 22

• https://doi.org/10.1016/j.oceaneng.2021.110390

OCP NONLINEAR PROGRAMMING PROBLEM (NLP) TRAJECTORY PLANNER 23(a)

OCP NONLINEAR PROGRAMMING PROBLEM (NLP) TRAJECTORY PLANNER 23(b)

TRAJECTORY PLANNER EVALUATION TRAJECTORY PLANNER (i) Comparison with an existing method – Validation (ii) Six scenario trajectories, with vs without speed reduction. Wind disturbances (UT up to ~0.75 m/s) applied throughout; solved with SQP in MATLAB– Simulation results (iii) Feasibility study: - Planner evaluation a) Grid-based simulations b) Stochastic simulation conditions 24

TRAJECTORY PLANNER - VALIDATION TRAJECTORY PLANNER Ship A 25

TRAJECTORY PLANNER - VALIDATION TRAJECTORY PLANNER Ship B 26

TRAJECTORY PLANNER – SIMULATION RESULTS Case 1 TRAJECTORY PLANNER Case 2 27

TRAJECTORY PLANNER – SIMULATION RESULTS Case 3 TRAJECTORY PLANNER Case 4 28

TRAJECTORY PLANNER – SIMULATION RESULTS Case 5 TRAJECTORY PLANNER Case 6 29

TRAJECTORY PLANNER – EVALUATION 1. Grid-based simulations TRAJECTORY PLANNER Simulation results 30

TRAJECTORY PLANNER – EVALUATION 2. Stochastic simulation conditions TRAJECTORY PLANNER Computation time 50% 62% 71% 75% initial recomputation 1 recomputation 2 recomputation 3 31

CONTRIBUTION 1 - TAKEAWAYS TRAJECTORY PLANNER Linear guess, single layer Prospective improvement/ future research focus Matches an expensive warm-started baseline ➢ What is the role of maneuvering model in while removing the global-optimization step. planner’s computation time? ➢ Besides control inputs guess, what other variables can be adjusted before Practical & robust Actuator limits, wind & ship domain included; 75% feasibility with cheap recomputation. recomputation? ➢ Does the safety envelop (ship domain) match actual ship navigation? Deceleration guidelines = safety Embedding empirical speed limits keeps the approach in the empirically recommended region. 32

PART 02 · CONTRIBUTION 2 A linear, low-speed maneuvering model 33

LOW-SPEED MODEL – OVERVIEW LOW-SPEED MODEL PROBLEM A PROBLEM B Nonlinear – considerably accurate but expensive Scale effects & narrow data The MMG model gives the planner accuracy but possibly Tank-test coefficients need uncertain model-to-full-scale increases computation cost — from Contribution 1. correction [16,17]; many low-speed models are extrapolated from too-narrow ranges [6,7]. R E SP ON SE Make the model deliberately simple — and identify it directly from full-scale data, so there is nothing to re-scale. 16. Michio Ueno, Yoshiaki Tsukada, and Yasushi Kitagawa. “Rudder effectiveness correction for scale model ship testing.” In: Ocean Engineering 92 (2014), pp. 267–284. issn: 00298018. doi: 10.1016/j.oceaneng.2014.10.006. 17. Michio Ueno and Yoshiaki Tsukada. “Rudder effectiveness and speed correction for scale model ship testing.” In: Ocean Engineering 109 (2015), pp. 495–506. issn: 0029-8018. doi: 10.1016/J.OCEANENG.2015.09.041. 34

SUBJECT SHIP LOW-SPEED MODEL A full-scale coastal ship Full-scale logs from this vessel ground both the low-speed model and the ship domain (Contribution 3). 35

MANEUVERING MODEL Linearize about hover, not zero speed LOW-SPEED MODEL The VecTwin rudder can redirect propeller thrust so net surge force is zero — the ship is stationary and can perform linearized motions, including crabbing (pure lateral translation) [18] . THE HOVER EQUILIBRIUM Rudder angles |δhover| Propeller n₀ Net surge force 70°–80° constant rps ≈0 Conventional Equilibrium at straight ahead motion; constant forward speed. This contribution Equilibrium at hover condition (stationary); zero forward 18. Dimas M Rachman, Yusuke Aoki, Yoshiki Miyauchi, Naoya Umeda, and Atsuo Maki. “Experimental low-speed positioning system with VecTwin rudder for automatic docking (berthing).” In: Journal of Marine Science and Technology 28.3 (2023), pp. 689–703. speed. 36

MANEUVERING MODEL LOW-SPEED MODEL Relationship between the two coordinate systems: Nonlinear Linear Hydrodynamic forces and moment: 37

MANEUVERING MODEL – TAYLOR EXPANSION LOW-SPEED MODEL First order Taylor expansion about the hover (initial equilibrium) condition: When hovering, are zero. Additionally, since the ship is stationary, By denoting the partial derivatives as we have: 38

MANEUVERING MODEL Wind-induced forces and moment: LOW-SPEED MODEL Rigid - body kinetics [19] 11. Toshifumi Fujiwara, Michio Ueno, and Tadashi Nimura. “Estimation of wind forces and moments acting on ships.” In: Journal of the Society of Naval Architects of Japan 1998.183 (1998), pp. 77–90. 19. Thor I. Fossen. “Handbook of Marine Craft Hydrodynamics and Motion Control.” In: (2011). doi: 10.1002/9781119575016 39

MODEL PARAMETERS IDENTIFICATION LOW-SPEED MODEL Data 40

MODEL PARAMETERS IDENTIFICATION LOW-SPEED MODEL Data (a) (c) (b) (d) 41

MODEL PARAMETERS IDENTIFICATION LOW-SPEED MODEL Training Data vs Testing Data States Control inputs 42

OPTIMAL MODEL PARAMETERS LOW-SPEED MODEL 43

MODEL TESTING AND VALIDATION LOW-SPEED MODEL Port 1 44

MODEL TESTING AND VALIDATION LOW-SPEED MODEL Port 2 45

MODEL TESTING AND VALIDATION LOW-SPEED MODEL Port 3 46

MODEL TESTING AND VALIDATION LOW-SPEED MODEL Port 4 47

MODEL TESTING AND VALIDATION LOW-SPEED MODEL Port 5 48

CONTRIBUTION 2 - TAKEAWAYS LOW-SPEED MODEL Hover-based formulation Full-scale, no scaling Linearizing about the VecTwin hover state — not straight- Identified directly from operational data — no tank tests, ahead motion — is a novel, physically apt reference for no model-to-full-scale correction. berthing. Simple, yet reliably accurate Ready for the loop Tracks full-scale motion closely across ports. Lightweight and linear — a promising candidate to replace the MMG model in the planner. 49

PART 03 · CONTRIBUTION 3 An adaptive ship domain 50

OVERVIEW SHIP DOMAIN • A ship domain is the keep-clear region around a hull [20,21]. • Analytical and AIS-based domains dominate existing PORTS CONSTRAIN IN TWO PLANES Vertical Shallow water limits under-keel clearance. research. Quays, buoys, leading lines & traffic • Existing research focus on open sea and restricted waters Horizontal bound the space — even with no moving obstacle. (channels) navigation. Objectives: i Environment-aware boundary Redefine navigable space with vertical (shallow-water) + horizontal (infrastructure) constraints. ii Anisotropic, speed-scaled domain Four semi-axes that differ from each other and vary with ship size iii Identify from full-scale data Identify & validate parameters on real berthing/unberthing records and speed. 20. Rafał Szłapczyński and Joanna Szłapczyńska. “Review of ship safety domains: Models and applications.” In: Ocean Engineering (2017). doi: 10.1016/J.OCEANENG.2017.09.020. 21. Ahmet Baran, Remzi Fışkın, and Hakkġ Kiźi. “A Research on Concept of Ship Safety Domain.” In: TransNav: International Journal on Marine Navigation and Safety of Sea Transportation (2018). doi:10.12716/1001.12.01.04. 51

SUBJECT SHIP SHIP DOMAIN A full-scale coastal ship Full-scale logs from this vessel ground both the low-speed model and the ship domain (Contribution 3). 52

SUBJECT SHIP SHIP DOMAIN Closest Point of Approach (CPA) 53

REDEFINED NAVIGABLE SPACE SHIP DOMAIN 54

REDEFINED NAVIGABLE SPACE SHIP DOMAIN Port navigation limit (red buoy) Starboard limit (green buoy) Obstruction (yellow) Other aids / leading lines The ship is measured against the operationally navigable harbor — not the raw geometry. 55

SHIP DOMAIN MODEL SHIP DOMAIN Model equations: where, Coordinates along the quaternion curve of the elliptical domain: where, Ship speed influence coefficient, unique for each axis. 56

MODEL PARAMETERS IDENTIFICATION SHIP DOMAIN Data acquisition At each point, distance to the nearest boundary (dCPA) is measured fore, starboard, aft, and port — four half-extents that become the domain's semi-axes. 57

MODEL PARAMETERS IDENTIFICATION SHIP DOMAIN Data 58

MODEL PARAMETERS IDENTIFICATION SHIP DOMAIN Training data vs Testing data 59

OPTIMAL MODEL PARAMETERS SHIP DOMAIN Ship domain at Port A 60

MODEL TESTING AND VALIDATION Ship domain at Port B SHIP DOMAIN Ship domain at Port C 61

MODEL TESTING AND VALIDATION Ship domain at Port D SHIP DOMAIN Ship domain at Port E 62

MODEL TESTING AND VALIDATION SHIP DOMAIN Analytical vs Empirical 63

CONTRIBUTION 3 - TAKEAWAYS SHIP DOMAIN Empirical, not analytical Speed-dependent & anisotropic Four semi-axes learned from 152 full-scale maneuvers Axes anisotropically expand and elongate with ship speed. across five ports. Environment-aware Implementation in the planner? Measured against a boundary that folds in depth & Redefine the safety envelop empirically — the collision infrastructure; pilot-bounded. limit in Contribution 1. 64

PART 04 Concluding remarks 65

SUMMARY CONTRIBUTION 1 CONCLUSION First berthing planner to embed empirical deceleration guidelines as constraints — and to show a linear guess replaces double-layered warm-starting. CONTRIBUTION 2 A low-speed model linearized about the hover condition, identified from full-scale data, with the parameter-cancellation problem handled explicitly. CONTRIBUTION 3 ∑ An empirical, adaptive, anisotropic ship domain for berthing — environment-aware and validated across ports. A coherent methodology for leveraging full-scale operational data for autonomous berthing algorithms. 66

LIMITATIONS CONCLUSION Not yet integrated end-to-end Planner edge cases The model & domain are validated separately — not yet run The linear guess still struggles to converge on some hard live inside a/the planner. configurations; recomputation mitigates, not eliminates. Model scope Domain size Identified for one ship — generalization needs model Current - one ship, five ports. Standardization require modifications and re-identification. validation across many ship types & environments. 67

FUTURE WORK CONCLUSION PLANNER LINEAR MODEL SHIP DOMAIN ➢ Adaptively tune the linear guess for ➢ Quantify the from full-scale effects ➢ Many ships & ports; find invariants complex and real port of trim and sinkage in ship motions; ➢ Quantify a unified safety criterion environments extend validity into shallow water. linking actuation capacity to ➢ Swap in the linear model for the MMG. ➢ Explore generalization across domain size. different ship types ➢ Compare analytical vs proposed empirical domain in trajectory planning 68

PUBLICATIONS CONCLUSION A practical and online trajectory planner for autonomous ships' berthing, incorporating speed control JOURNAL Agnes N Mwange, Dimas M Rachman, Rin Suyama, and Atsuo Maki. In: Journal of Marine Science and Technology (2025), pp. 238 – 254.DOI: 10.1007/S00773-025-01048-0 Quantitative evaluation of full-scale ship maneuvering characteristics during berthing and unberthing JOURNAL Agnes N. Mwange, Yoshiki Miyauchi, Taichi Kambara, Hiroaki Koike, Kazuyoshi Hosogaya, Atsushi Ishibashi, and Atsuo Maki. In: Journal of Marine Science and Technology (2025). ISSN: 09484280. DOI: 10.1007/s00773-025-01098-4. Online Trajectory Planner that Enhances Berthing Safety through Incorporating Speed Reduction Guidelines CONFERENCE Agnes N. Mwange, Dimas M. Rachman, and Atsuo Maki. In: Conference Proceedings The Japan Society of Naval Architects and Ocean Engineers 37 (2023), pp. 61 – 63. Development and Identification of a Linear Low-Speed Ship Maneuvering Model from Full-Scale Data SUBMITTED TO BE SUBMITTED Agnes N. Mwange, Taichi Kambara, Kouki Wakita, Kazuyoshi Hosogaya, and Atsuo Maki. 2026. arXiv:2607.01739 [eess.SY]. url: https://arxiv.org/abs/2607.01739. Empirical Modeling of a Ship Domain for Berthing and Unberthing Using Full-Scale Data Agnes N. Mwange, Kazuyoshi Hosogaya, Atsushi Ishibashi, and Atsuo Maki. 69

• https://arxiv.org/abs/2607.01739

DOCTORAL DISSERTATION DEFENSE Thank you… THE UNIVERSITY OF OSAKA · JULY 2026