Add Figure 5 Ant flat-ground and maze experiments

README.md

@@ -6,39 +6,50 @@ tags:
 - reinforcement-learning
 - drosophila
 - simulation
-pretty_name: FlyControl paper records — Figures 2–4 and S3–S6
+pretty_name: FlyControl paper data
 ---
-# FlyControl — paper data and provenance
+# FlyControl data
 
-This is a private, author-review snapshot for **A Drosophila connectome prior enables adaptable embodied control**. It covers Figure 2, Figure 3, Figure 4, and Supplementary Figures S3–S6 from the final manuscript supplied on 2026-11-02. Figure 5 and Figure 6 are outside this snapshot.
+Research records for **A Drosophila connectome prior enables adaptable embodied control**.
 
-The manuscript identity is SHA-256 `cb7dac9a2c20305f6228c98766d7d0f361ce59c058e1e13b88bee59d38152a30`. The manuscript itself is not distributed here. The corresponding code is [FlyControl](https://github.com/THU-NeuroML/FlyControl/tree/1a49c33bf1e6eab17cb2953123e02ecc2549a5f4), revision `1a49c33bf1e6eab17cb2953123e02ecc2549a5f4`.
+Code: [FlyControl](https://github.com/THU-NeuroML/FlyControl/tree/1a49c33bf1e6eab17cb2953123e02ecc2549a5f4).
 
-Seven figure-specific archives contain numerical source tables, included episode results and traces, mechanism-analysis records, exact derived graphs, effective protocols, and the visual source assets needed by the main-figure panels. Large collections of small files are bundled without changing their contents. Checkpoints and loading instructions are in [FlyControl-models](https://huggingface.co/UncleStrawberryBear/FlyControl-models).
+| Figure | Experiment | Models |
+|---|---|---|
+| Figure 2 | Fly terrain locomotion | Adult, MLP and ER; 20 seeds each |
+| Figure 3 | Fly turning control | Adult and MLP; 24 seeds each |
+| Figure 4 | Fly speed control | Adult and MLP; 20 seeds each |
+| Figure 5b–d, S7 | Ant flat ground and floor friction | Adult and MLP; 20 seeds each |
+| Figure 5e | Ant U-Maze, Medium Maze and Large Maze | Seven controller families; 20 seeds per maze and family |
+| Figure 5f–j | Ant Medium Maze trajectories and navigation | Adult and MLP; 20 seeds each |
+| S3 | Developmental direction control | Adult and Larva; 6 seeds each |
+| S4 | Developmental speed control | Adult and Larva; 6 seeds each |
+| S5 | BANC descending and motor readouts | 6 seeds per readout; initial and final states |
+| S6 | BANC regional recovery | 3 seeds per readout |
+| S8 | Ant Large Maze trajectories, speed and route complexity | Seven controller families; 20 seeds each |
 
-| Paper item | Evidence / task | Checkpoint files |
-|---|---|---:|
-| Figure 2 | Terrain; Adult, MLP and ER, 20 seeds each | 60 |
-| Figure 3 | Turning; Adult/MLP, 24 seeds each, plus 24 Adult initial states | 72 |
-| Figure 4 | Speed; Adult/MLP, 20 seeds each | 40 |
-| S3 | Developmental direction; Adult/Larva, 6 seeds each | 12 |
-| S4 | Developmental speed; Adult/Larva, 6 seeds each | 12 |
-| S5 | BANC descending/motor readouts, 6 seeds each, initial and final | 24 |
-| S6 | BANC regional recovery, 3 seeds per readout | 6 |
 
-`archive_index.json` records each archive's size, SHA-256 and member count. `manifests/<figure>.jsonl` records every extracted path, size, SHA-256 and original local source. `figure_map.csv` maps the final paper numbering to historical directories: final S5/S6 correspond to historical `s06`/`s07`. No standalone S1/S2 archive is included in this requested snapshot.
+The seven Ant maze controller families are Adult connectome, MLP, GRU, LSTM, Transformer, Reservoir and Reservoir+RO. The Ant flat-ground MLP uses two 1,378-unit ReLU hidden layers.
 
-## Contents and statistical units
+The archives contain episode records, trajectories, figure source tables, statistical summaries and experimental inputs. Corresponding policies are in [FlyControl-models](https://huggingface.co/UncleStrawberryBear/FlyControl-models).
 
-- Figures 2–4 retain the plot inputs, episode/seed tables, full comparison-family summaries, perturbation grids and selected mechanism histories. Figure 2 also includes the final corrected axis/surface and ER comparison tables. Figure 3 preserves the broader registered comparison data while checkpoint coverage is restricted to the plotted Adult/MLP cohort and the necessary initial Adult states.
-- S3 contains direction episodes and the full 22-comparison seed-level inference family, including the jointly corrected terrain endpoints. These supplemental policies are distinct from the main Figure 3 cohort.
-- S4 contains the full speed endpoint/seed tables and all 240 stored step/sine traces used by the displayed developmental analyses. Other rows in the endpoint table can reference historical trajectories outside this selected trace subset; `plot_trace_manifest.csv` enumerates the included traces.
-- S5 contains before/after BANC class sensitivity, per-window/per-state records, neuron annotations, and the 32 exact observed common histories. It does not require publishing all policies that originally generated those histories.
-- S6 contains all 1,890 analyzed stable recovery branch traces, 54 matched branch snapshots, unpushed neural references, the probe records, regional masks/edge fractions, and recorded quality-control summaries. The six recovery policies are supplied in the model repository.
+## Figure 5, S7 and S8
 
-Training seed identifies the policy-level unit; evaluation seed identifies a repeated rollout or reset. Do not treat repeated trajectories as independent trained models. Preserve failure flags, partial durations and missing windows. Original raw and multiplicity-adjusted statistics are retained; use the stated family rather than recomputing correction over only significant/displayed rows.
+`archives/figure5.tar.gz` contains:
 
-Column names retain the original units: `_s` seconds, `_mm` millimetres, `_mm_s` millimetres per second, `_rad` radians, `_deg` degrees, and `_hz` hertz. Statistics tables may contain an explicit `unit` field that takes precedence. Blank/NaN fields can mean unavailable or inapplicable measurements and must not be replaced by zero. NPZ contents and protocol JSON files retain the original task-specific variable names; this is a research archive, not a uniform supervised-learning table.
+- Ant flat ground: 4,000 evaluation episodes, motion endpoints and displacement trajectories from 40 policies.
+- Floor friction: 4,000 episodes across four friction coefficients, with seed summaries for S7.
+- Ant mazes: 42,000 evaluation episodes from 420 policies across U-Maze, Medium and Large mazes.
+- Medium Maze: example trajectories, deadline curves, geodesic efficiency, route-turn groups and physical-failure records.
+- Large Maze: seven example trajectories and speed records, route-stratified metrics and seed summaries for S8.
+- Statistical source tables, including the 26- and 29-comparison families.
+- Teacher-labelled training samples and an index mapping them to maze, controller, optimization seed and training stage.
+
+Extracted files are under `data/figure5/`. Start with `experiment_map.csv`, `model_index.json`, `maze/episodes.csv`, `source_data/`, `S7/`, `S8/` and `training_labels_index.json`. Identical training-label files are stored once and referenced by checksum.
+
+Maze training labels were generated using the GoalAnt teacher from [Farama-Antmaze-Dataset-Generation](https://github.com/LukaszSawala/Farama-Antmaze-Dataset-Generation/tree/fb253c9efbe2fa5ec256a6fc3f983ba33192d706), checkpoint SHA-256 `1b835cc5185756c5af8cd0720dae0e4895efcec48ee577c450d4a5a633fb953e`.
+
+Optimization seed identifies the policy replicate; reset seed identifies an evaluation episode. Trajectory distances are in metres, time in seconds and velocity in metres per second. Each source table retains its original field names and units.
 
 ## Download and extract
 
@@ -46,18 +57,12 @@ Column names retain the original units: `_s` seconds, `_mm` millimetres, `_mm_s`
 from huggingface_hub import snapshot_download
 data = snapshot_download(
     'UncleStrawberryBear/FlyControl-data', repo_type='dataset',
-    allow_patterns=['README.md', 'archive_index.json', 'manifests/figure2.jsonl',
-                    'archives/figure2.tar.gz', 'extract_data.py'])
+    allow_patterns=['README.md', 'archive_index.json', 'extract_data.py',
+                    'manifests/figure5.jsonl', 'archives/figure5.tar.gz'])
 ```
 
 ```bash
-python /path/to/data-snapshot/extract_data.py   --figure figure2 --output ./paper-data
+python /path/to/data-snapshot/extract_data.py --figure figure5 --output ./paper-data
 ```
 
-The extractor verifies the archive and every member checksum, rejects unsafe members, and preserves the recorded relative layout. Multiple figure archives can be extracted into the same directory. See `data/figure2/source_data`, `data/figure3/source_data`, and `data/figure4/source_data` for main-figure source tables. Supplemental locations appear in `figure_map.csv`; additional raw S5/S6 records live under `research/outputs/nmi_unified/`.
-
-Historical absolute paths inside original CSV/JSON records are provenance, not required machine locations. Use manifest paths after extraction; match trajectories or models by their SHA-256 identities. `source_aliases.jsonl` maps additional known original supplementary paths to included archive members. `plot_trace_manifest.csv` supplies the portable S4 trace paths. A historical path to an omitted exploratory run does not imply that run belongs in this snapshot.
-
-The archives retain exact FlyWire/Larva/BANC derived inputs; the upstream version manifests under `external_inputs/` identify their public sources. Use BANC release 626, as verified by the code's reproduction notes, rather than inferring a release from historical folder names. See `RIGHTS.md`.
-
-This is a private archival upload. Hash and coverage checks do not constitute a new full simulator reproduction of the paper. No DOI or public-release claim is made.
+Use `figure2`, `figure3`, `figure4`, `S3`, `S4`, `S5` or `S6` to extract the other archives. `archive_index.json` lists archive sizes and checksums; `manifests/` lists member paths, source paths and SHA-256 values. The extractor verifies these checksums. `figure_map.csv` maps paper figures to their archived records. Source and licensing information is in `RIGHTS.md`.

archive_index.json

@@ -6,7 +6,10 @@
     "S3",
     "S4",
     "S5",
-    "S6"
+    "S6",
+    "figure5",
+    "S7",
+    "S8"
   ],
   "archives": [
     {
@@ -57,6 +60,18 @@
       "bytes": 3406429927,
       "sha256": "7717995396c90d3da544aa0c510a46b918a8846ed834c006e2b06147ce6d8ab0",
       "members": 2092
+    },
+    {
+      "paper_item": "figure5",
+      "includes": [
+        "Figure 5",
+        "Figure S7",
+        "Figure S8"
+      ],
+      "path": "archives/figure5.tar.gz",
+      "bytes": 974876471,
+      "sha256": "f15bd986490ef05016d166c0c6995161d74c599ca16c8d32335e69c927eb3440",
+      "members": 1709
     }
   ]
 }

archives/figure5.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f15bd986490ef05016d166c0c6995161d74c599ca16c8d32335e69c927eb3440
+size 974876471

figure5_validation.json

@@ -0,0 +1,16 @@
+{
+  "maze_models": 420,
+  "maze_episodes": {
+    "umaze": 14000,
+    "medium": 14000,
+    "large": 14000
+  },
+  "flat_models": 40,
+  "flat_native_episodes": 4000,
+  "friction_episodes": 4000,
+  "unique_training_label_files": 162,
+  "archive_members": 1709,
+  "archive_verified": true,
+  "all_member_checksums_verified": true,
+  "S8_replay_models_verified": 7
+}

figure_map.csv

@@ -6,3 +6,6 @@ S3,archives/S3.tar.gz,S3/,12,data/supplementary/development
 S4,archives/S4.tar.gz,S4/,12,data/supplementary/development
 S5,archives/S5.tar.gz,S5/,24,data/supplementary/s06
 S6,archives/S6.tar.gz,S6/,6,data/supplementary/s07
+figure5,archives/figure5.tar.gz,figure5/,460,data/figure5
+S7,archives/figure5.tar.gz,figure5/flat/,40,data/figure5/S7
+S8,archives/figure5.tar.gz,figure5/large/,140,data/figure5/S8