Forum

[Skipper Don]

SafeHelm Concept — Open Source Collision Avoidance for d3kOS

Why This Matters: The Numbers

251 vessel collisions globally in 2024 (Allianz Global Corporate & Specialty)
78.8% of EU marine casualties involved human error (EMSA 2025)
2,659 marine casualties & incidents in EU waters in 2024 (EMSA)

Collision is the #1 cause of ship-related fatalities and #2 cause of all marine incidents globally. Nearly 80% trace
back to human error: misjudged encounters, watchkeeping mistakes, COLREGs misinterpretation.

An autonomous collision avoidance system that reliably handles these scenarios doesn’t just add convenience. It
targets the dominant cause of deaths and losses at sea.

The Core Problem: Why No Open Source Solution Exists

Despite 50 years of COLREGs and decades of collision avoidance research, no open-source, certifiable, production-ready
system exists.

The reason? COLREGs was written for humans, not algorithms.

Rule 2 — “Good seamanship” & “ordinary practice of seamen”
→ Entirely undefined. Cannot be mathematized. Requires simulating accumulated professional judgment.

Rule 5 — “Proper look-out by all available means”
→ Written assuming human eyes + hearing. No definition covers what a proper multi-sensor autonomous look-out is.

Rule 8 — “Ample time”, “take early and substantial action”, “keep well clear”
→ All subjective. How early is “early”? How large is “substantial”? An algorithm either over-reacts or under-reacts.

Rules 13–17 — Overtaking / Head-on / Crossing
→ Borderline encounters (head-on vs crossing) cause deadlock. Case law (Evergreen Marine v. Nautical Challenge) shows
even human courts disagree.

Multi-vessel scenarios — Three or more vessels converging
→ COLREGs was designed for pairwise encounters. No rule defines what happens when complying with one vessel forces
non-compliance with another.

The research verdict (Wróbel et al., 2022): “All existing collision avoidance methods are at best only partly
COLREGs-compliant.”

Four Technical Walls

1. Real-time performance on modest hardware
Deep RL and MPC algorithms work in simulation but are computationally heavy. Raspberry Pi has a 100ms budget. Research
code doesn’t optimize for deterministic timing or latency constraints.

2. Non-compliant vessels
COLREGs assumes the other vessel will comply. In practice, fishing boats, leisure craft, and high-traffic straits
routinely violate rules. A system that blindly obeys COLREGs when the other vessel doesn’t can actually increase
collision risk.

3. Sensor fusion under uncertainty
AIS can be absent, delayed, or spoofed. Radar needs ARPA processing. Cameras fail at night. Lidar is expensive. Real
systems must fuse imperfect streams and still make safe decisions in real time.

4. Explainability & certification
Maritime law requires auditable, explainable decisions. DRL approaches show best performance but are black boxes — you
cannot prove why a maneuver was taken. Flag states and classification societies cannot certify what they cannot
understand.

—
SafeHelm: A Design Proposal

We’re proposing SafeHelm — a collision avoidance module designed to address these gaps. This is a design concept. No
code has been written. No testing has been done. We’re asking the community: is this a problem worth solving? And how
would you solve it?

What SafeHelm Would Be (If Built)

A seven-layer architecture:

1. Sensor Fusion & Reliability — AIS + camera + GPS + IMU with confidence scoring. Explicit degraded modes (AIS-only,
camera-only, GPS-only).
2. Real-Time Execution Core — 100ms deterministic loop. Watchdog timers. Deterministic timing guarantees.
3. Vessel Dynamics Model — Turn radius, stopping distance, prop walk, wind drift, current correction. Not point-mass
physics — real boat physics.
4. COLREGs Decision Engine — CPA/TCPA calculations. Situation classification (head-on, crossing, overtaking).
Explicitly mark ambiguous cases rather than guess.
5. Multi-Vessel Conflict Solver — Priority graph, deadlock detection. When complying with one vessel forces
non-compliance with another, escalate to captain.
6. Human Interface — Voice alerts, AvNav widget, alarm budget (prevent fatigue). One-tap execute. Always overridable.
7. Safety Envelope — Hard constraints. Control barrier function. Never execute an unsafe command even if the planner
says to.

Open Questions We’d Ask the Community

On COLREGs Compliance:
– When SafeHelm encounters an ambiguous situation (head-on vs. crossing borderline), should it ask the captain rather
than decide?
– For multi-vessel scenarios where one rule conflicts with another, is escalation to the helm the right answer?

On Non-Compliance:
– Have you encountered fishing boats or leisure craft that don’t follow COLREGs? How would you want the system to
react?
– Should the system learn non-compliance patterns (e.g., this boat never turns despite being give-way) over time?

On Sensors:
– Would camera-detected targets (as virtual AIS entries) be trusted on your chart?
– What sensor failures have left you most exposed?

On Explainability:
– Would you deploy a system that can tell you why it recommended a maneuver (citing the specific COLREGs rule)?
– Would you trust it if it couldn’t?

On Trust:
– Would you want a [SUGGEST MANEUVER] button that you confirm, or [EXECUTE MANEUVER] that needs one tap?
– What would change your mind either way?

Why Open Source Matters

– No vendor lock-in. You own the code on your boat.
– Academic and maritime community can audit it against COLREGs and research.
– Community-driven. Boaters hit edge cases researchers miss.
– Free. Forever.

Tell Us What You Think

If this problem resonates with you:

– Reply in this thread. Does collision avoidance automation matter for your boating?
– Share a near-miss. A real encounter where you’d have wanted extra eyes or a second opinion.
– Ask hard questions. This design has gaps. Point them out.

—
Sources:
– Allianz Global Corporate & Specialty. 2024 Claims Review.
– EMSA (European Maritime Safety Agency). 2025 Statistics.
– Wróbel, K., et al. (2022). Autonomous Ships and New Challenges in Maritime Safety. WMU Journal of Maritime Affairs.
– IMO. International Regulations for Preventing Collisions at Sea, 1972 (COLREGs).

[Author]

Posts