docs: alternative ENC→costmap architecture (clean-room)#24
Draft
rolker wants to merge 5 commits into
Draft
Conversation
Captures a clean-room design exercise for generating nav2 costmaps from S-57 ENC data, produced without consulting the existing implementation, to be used as a comparison reference. Records inputs/constraints, four design questions (SOUNDG/DEPARE, UNSARE, regulatory, operating area), the multi-leg mission scenario, and the resulting layered architecture with explicit deferrals. Refs #23 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Round 2 additions to the clean-room design, prompted by follow-up questions on cold-start time, restart downtime, and overhead clearance. Architecture doc: - New Overhead layer (BRIDGE / CBLOHD / CONVYR) with VERCLR + vessel air-draft. Tide-conditional lethal, mirroring Depth with opposite sign. - New section 5.10 on persistence and lifecycle: persistent feature-index cache, cold-start estimates (~2-4 s harbor / ~8-20 s regional with cache, ~30-90 s without), restart downtime implications, and a deferred out-of-process ENC service alternative. - v1 scope updated; new deferrals (drawbridge state, EMI hazards, out-of-process service). - Layer count is now six (was five); subsections 5.6-5.9 renumbered. New design_evolution.md tracks revision history and rationale, including the honest accounting of what was missing before this round (overhead clearance was a real gap; caching was implicit, not designed; air-draft was listed but unused). Refs #23 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a standalone neutral summary of Samuel Reed's 2018 MS thesis "Providing Nautical Chart Awareness to Autonomous Surface Vehicles" (UNH ECE, advised by Brian Calder and Val Schmidt). The work in this repo is the spiritual successor to Reed's research, and many of the same problems (DEPARE/SOUNDG conflict resolution, WATLEV qualitative depth handling, tide-corrected charted depth, vessel-size buffering, ENC scale and uncertainty) recur in any ENC-driven autonomy system. The summary captures Reed's MOOS-IvP architecture, Depth-Based A* mission planner with novel cost function balancing channel-following against shortest-path, threat-level encoding (TL 0-5), ENC_DB / ENC_Contact / BHV_ENC_OA reactive stack, MOOSTides tide handling, field test results on EchoBoat and C-Worker 4, and Reed's stated limitations (reactive local minima at multi-obstacle scenarios; planner close approaches to point obstacles in deep water; ENC cartographic scale issues including the breakwater installed 2006 but omitted from ENC until 2018, and the "Breakers" label located 88 m from the actual hazard). The doc is descriptive only; no comparison against this repo's existing implementation or against the alternative-architecture proposal in this PR. Comparison is a separate planned step. design_evolution.md updated with Round 3 entry. Refs #23 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reed publishes under github.com/sji367 with three public repos that
match the thesis components:
- sji367/MOOS_ENC: the OA stack (BHV_OA, BHV_OA_poly, ENC_converter,
ENC_Search, ENC_WPT_check, AOF_Gauss). The published code splits the
thesis's single BHV_ENC_OA into separate point/polygon behaviors.
- sji367/moos-ivp-reed: moos-ivp-extend scaffolding for the work.
- sji367/ENC_Mission_Planner: Qt5 offline mission planner.
valschmidt/encgrid is Val Schmidt's standalone C++ port of the
gridding pipeline, with notes_on_encgrid.txt surfacing implementation
details not in the thesis prose: MULTIPOINT (SOUNDG) bypass, DEPCNT
special-case linestring handling, OBSTRN/PONTON/FLODOC/DYKCON
polygon-conversion path, internal cm-scaling, geotiff orientation.
Reed's repos are essentially frozen since his thesis defense; "last
push" search timestamps are metadata changes, not code commits.
Workspace connections (verified via git log, not inferred from
filename overlap):
- camp/src/camp/astar.h was imported from Reed on 2020-02-20 (commit
f464452, "Add initial AStar support for Sam Reid's work"), about
14 months after the thesis. CAMP itself started 2016-10-25 by
Roland Arsenault, predating Reed's ENC_Mission_Planner (2017-05-16)
by seven months. Shared filenames between the two Qt projects
exist; direction of influence on those files is not established.
- s57_tools was first committed 2021-06-25 by Roland, with no port
history from encgrid or moos-ivp-reed. The conceptual overlap is
genuine but the code is independent.
design_evolution.md updated with Round 4 entry that includes a
correction note on an earlier draft of this round which had inferred
lineages from shared filenames; the user flagged it and existing
memory feedback was reinforced ("git log first" before claiming code
lineage).
Refs #23
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User's alternative design idea, captured faithfully without analysis or padding, ready to be picked up in the next discussion. The idea models the way a human mariner plans a voyage: assemble a per-voyage collection of ENC cells at varying scales, fine at departure and arrival, coarse during transit. The planner traverses the collection by jumping to a coarser or finer chart as the boat's position requires. Cost can be supplied to nav2 as either per-chart rasterized costmaps or vector-direct queries (skipping rasterization entirely). Lists open questions for the next discussion: ENC scale metadata mapping, chart-boundary handling, where the chart-jump decision lives in the planner pipeline, the right nav2 plugin point, vector-direct planning effort, no-chart-covers-gap recovery, and tide/dynamic-state behavior across scales. The repo now has three distinct designs on file (clean-room costmap architecture, Reed prior art, chart-scale planner), positioned for the eventual comparison exercise. design_evolution.md updated with Round 5 entry. Refs #23 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
docs/design/alternative_costmap_architecture.md(749 lines).produced without consulting the existing implementation in this repo.
(SOUNDG/DEPARE, UNSARE, regulatory, operating area), the multi-leg mission
scenario, and the resulting layered architecture — alongside the
conclusions, with explicit "what's deferred" and "where confidence is low"
sections.
The intent is to compare the two designs in a follow-up. This PR just lands
the alternative; the comparison is out of scope here.
Test plan
s57_grids/s57_layer/marine_charts.Closes #23.
Authored-By:
Claude Code AgentModel:
Claude Opus 4.7 (1M context)