Add Figure 5 Ant flat-ground and maze experiments

README.md

@@ -9,51 +9,56 @@ tags:
 datasets:
 - UncleStrawberryBear/FlyControl-data
 ---
-# FlyControl — archived paper policies
+# FlyControl models
 
-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).
 
-The repository contains **226 original PyTorch checkpoint files** (190 final policies and 36 initial states), their effective configuration files, `model_index.json`, and a small loading helper. Files retain their original bytes, graph buffers, and saved normalization/training state. The total checkpoint size is approximately 11.01 GiB.
+| 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 |
 
-These are task-specific simulator policies. Main-figure checkpoints and supplemental checkpoints use different numerical engines. In particular, S5 speed and S6 recovery policies are distinct despite overlapping seed labels. The index records original paths and SHA-256 identities to disambiguate historical batches. For S4, Adult seed 3 uses the recorded `seed3_attempt1` policy. The S3 larval checkpoints use a dense recurrent graph; S4 larval checkpoints use sparse edges.
+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.
 
-## Download and load
+The repository contains 686 checkpoint files: 650 final policies and 36 initial states. Figure 5 contributes 40 flat-ground policies and 420 maze policies. Checkpoints include actor and critic weights, saved observation normalization, and the graph, goal adapter and goal router used by the corresponding policy.
+
+`model_index.json` identifies each checkpoint by figure, experiment, controller, seed, configuration and SHA-256. Figure 5 entries are also listed in `figure5/model_index.json`. Model IDs follow `figure5/<flat|umaze|medium|large>/<method>/seed_<0–19>/final`.
 
-Use an authenticated Hugging Face account with access. Install the dependencies specified by the pinned FlyControl code, then download the model snapshot and the matching data archive:
+## Download and load
 
 ```python
 from huggingface_hub import snapshot_download
 models = snapshot_download('UncleStrawberryBear/FlyControl-models')
 ```
 
-For a single cohort, use `allow_patterns` to select its checkpoints/configs together with `model_index.json` and `load_policy.py`. Main-figure policies require the matching graph in the extracted [data repository](https://huggingface.co/datasets/UncleStrawberryBear/FlyControl-data). Extract that figure archive into an empty `paper-data` directory.
+Install the dependencies from the linked FlyControl revision. Load a Figure 5 policy with:
 
 ```bash
-git clone https://github.com/THU-NeuroML/FlyControl.git
-git -C FlyControl checkout 1a49c33bf1e6eab17cb2953123e02ecc2549a5f4
-python /path/to/model-snapshot/load_policy.py   --model-id figure2/connectome/seed_2011/final   --code-dir ./FlyControl --data-dir ./paper-data
+python /path/to/model-snapshot/load_ant_policy.py \
+  --model-id figure5/medium/adult/seed_0/final \
+  --code-dir /path/to/FlyControl
 ```
 
-This performs a strict state load and a numerical forward pass on zero input; it does not start a simulator. Supplemental policies reconstruct their graph from the saved checkpoint and do not need `--data-dir` for this numerical check. Run the helper in a fresh Python process so that another checkout's `fly_control` or `larva` modules cannot shadow the pinned code.
-
-The helper returns the policy and saved observation normalization state. Inputs must follow the task's original observation construction and normalization. Figure 2 uses 688 inputs (four 172-channel frames) and 6 actions; Figure 3 uses 64 inputs and 2 actions; Figure 4 and S3–S6 use 51 inputs and 2 actions. Main-figure Gaussian coordinates and supplemental action semantics are interpreted by their respective simulator engines; a raw network output is not a physical motor command.
+The helper verifies the checkpoint, constructs its policy, restores the saved state and runs a numerical forward pass. Its Python function `load_ant_policy` returns the policy and observation-normalization statistics. Figure 5 policies use 109 observation channels and eight action channels. Evaluation settings, including action scale, smoothing and waypoint planner, are stored in each index entry.
 
-Legacy initial checkpoints contain Python training state, so the helper uses `torch.load(..., weights_only=False)` on these known, checksum-verified research files. Original configurations and checkpoints retain historical filesystem paths as provenance; they are not portable filesystem instructions. Map identities through `model_index.json` and use a new run directory with the code's task interfaces.
+For Figures 2–4 and S3–S6, use `load_policy.py`. Figures 2–4 use the matching graph from the data archive:
 
-## Validation and boundaries
-
-All 226 files were copied without transformation and verified against source SHA-256 values (and historical recorded hashes where available). Every policy tensor was checked for finite values. Eighteen representatives covering every figure/method/stage combination passed strict state loading, exact state equality, and three finite forward steps with the pinned code. Main-figure checkpoint budgets were checked: 32,768 / 8,192 / 15,872 steps for Figures 2 / 3 / 4. Details are in `validation.json`.
+```bash
+python /path/to/model-snapshot/load_policy.py \
+  --model-id figure2/connectome/seed_2011/final \
+  --code-dir /path/to/FlyControl --data-dir /path/to/paper-data
+```
 
-This upload did not retrain policies or rerun full simulator evaluations. Historical source data are provided separately; the numerical loading checks do not certify a fresh end-to-end reproduction of all manuscript results. See `RIGHTS.md` for the current private-review licensing status.
+Data and trajectories: [FlyControl-data](https://huggingface.co/datasets/UncleStrawberryBear/FlyControl-data). File and loading checks are recorded in `validation.json` and `figure5/validation.json`. Source and licensing information is in `RIGHTS.md`.

checkpoints/figure5/flat/adult/seed_0/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb0449e58e7bee13147681acdd615cd1fd1e77a1ea82f2403372fdc0ea0662b9
+size 100764589

checkpoints/figure5/flat/adult/seed_1/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a979efcea20cb32ede83f6c01a77ff42b13a97a896bac284c6383ea73d2013e
+size 100689197

checkpoints/figure5/flat/adult/seed_10/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c425d06a2400dfd52aac230e42f1c9b3d490d3f324da56d6a798f2923ae04772
+size 100697133

checkpoints/figure5/flat/adult/seed_11/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff34f91b830ade9702434280294257bceb8abac195a38f0064f4cfd651361610
+size 100736301

checkpoints/figure5/flat/adult/seed_12/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ec68d266f9e3c0fa2850c2d8ce57f458911830d1f5e90433f328b5cb75f514f
+size 100609005

checkpoints/figure5/flat/adult/seed_13/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5da3e9feb67ce1ea8bccaa51c3c46fa79a4cdc04507d84fdb8ced80da53e353
+size 100772397

checkpoints/figure5/flat/adult/seed_14/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dea93e510fa13db9368bc1c119433056946251bc53a5e846181798bad0eb8b13
+size 100689197

checkpoints/figure5/flat/adult/seed_15/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f016d78e96dbffcc5f253ef454e5ab181f9983349ff60ba6eff069c8e5b8987
+size 100809069

checkpoints/figure5/flat/adult/seed_16/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6da21c02f85799ba4d7819e25c11ff345b1e3507bd3be0b35f41180dff4e2647
+size 100836845

checkpoints/figure5/flat/adult/seed_17/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71fce42684357ed6255a3d28b93c2ede12ea021662e62d671da7f866acf2de2c
+size 100693229

checkpoints/figure5/flat/adult/seed_18/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55481eafbeff5c69a16e7713c7315c0a325ce8f3ba19450811bf2f02bfc7189f
+size 100808621

checkpoints/figure5/flat/adult/seed_19/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0d9d4fc424ef569d6bd7532882a2fca6910e9bfc4f53c88999ebf056ef81198
+size 100786733

checkpoints/figure5/flat/adult/seed_2/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25625924462cbb62d391ccd20a337f14274840fe06d618448d140d6898685087
+size 100602093

checkpoints/figure5/flat/adult/seed_3/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7749e880023297590f87fbbc60f0f3764444003323e9dc40e406b5fa435c959
+size 100793133

checkpoints/figure5/flat/adult/seed_4/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8810277633e212f14f3736a76f9273f5d35521e88403fb568957d474957f5172
+size 100617069

checkpoints/figure5/flat/adult/seed_5/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89f9eb1898cf5bada7c5450f1028da6e6c03927c03a4a1d86e1631573386b125
+size 100604781

checkpoints/figure5/flat/adult/seed_6/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04c953ef3727ea854d896d9afa3dd5b9d6d37df1442227f147024cf54b04997f
+size 100801069

checkpoints/figure5/flat/adult/seed_7/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5edf37fab2e7ef3b5ecdce91965d1e07cb6f0e71ae39240e3d9510dc05ecfdcd
+size 100603821

checkpoints/figure5/flat/adult/seed_8/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85623ccb8aaf4180cb47f58019de6bec3b1f26ce60609548dc640716baf9d916
+size 100775917

checkpoints/figure5/flat/adult/seed_9/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fda334a8ed66ae7996d3c5f6a93253a2bbe81fb982bf111d7ea5d8b06ae5044
+size 100635181

checkpoints/figure5/flat/mlp/seed_0/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:238c763b5fb7daee4a0e88eee247a29a30cca8ee8f020c11d74ae7aadfe4cfed
+size 26098061

checkpoints/figure5/flat/mlp/seed_1/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86e500df1423dec0ba0b76d2bc6fda6867f5d481f8e9b971c5188661966c3979
+size 26103565

checkpoints/figure5/flat/mlp/seed_10/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13c508f9037c4ff76ab7f799dc850205a6fe4619f35aaa986c5efc2e63cf43bc
+size 26149517

checkpoints/figure5/flat/mlp/seed_11/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70436c21549034cbdfa2b1dfaac656180ad7e3ddbba9a829969edb4509a9be84
+size 26034509

checkpoints/figure5/flat/mlp/seed_12/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea57791966424c7f20c6611ac4fdf93b5ee18a3f0b5fa40eee2285b50a454f41
+size 26177677

checkpoints/figure5/flat/mlp/seed_13/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:331c46c00f0c36f839881fe55443daaedc6327679a367de11353b9d0c7fa721f
+size 26142541

checkpoints/figure5/flat/mlp/seed_14/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:162a78494a2f482033913cdbc1d598fc5c00926f5709caf3a91ff3ac2958c946
+size 25940749

checkpoints/figure5/flat/mlp/seed_15/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29eea20b848b66dc84a5aeb4c02209b64b4f7ad1c9079cdfa0e85ba2074da53f
+size 26114125

checkpoints/figure5/flat/mlp/seed_16/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19b5517d0b12c3f2afc145c496276a0cfead1bb536396e7a6295c89760456e79
+size 26185485

checkpoints/figure5/flat/mlp/seed_17/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fde0e5bf6f4209e542ec0aac6604895b56cbf60b442a1686c1056c05e5036faf
+size 26063309

checkpoints/figure5/flat/mlp/seed_18/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66262658ce43bf35d52f59c4866c1c13a0bf57f4dce788aac82375486d0a757d
+size 26101133

checkpoints/figure5/flat/mlp/seed_19/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bd855dd243fedda9fba86713ee173ba74779ccf7e0e9febb4ad61e464aa7088
+size 26133261

checkpoints/figure5/flat/mlp/seed_2/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0540090e78472516ee7e9ef57a1a0345037c78b6f49183983faa4eb3dfb1e3d9
+size 26156173

checkpoints/figure5/flat/mlp/seed_3/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:851cf16614a08d6cccbc69a0ca7fd428ad90261842ec9945672862116895c92a
+size 25941005

checkpoints/figure5/flat/mlp/seed_4/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67cc9b04c0968c135ce0a3cb6b9263044959c0164a2f3ba24740a2be274459b6
+size 26205005

checkpoints/figure5/flat/mlp/seed_5/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3cbd6a61bd9f3653217889ba1b6e596e7d4cbd3783a6c29df047c50ca5a44124
+size 25981069

checkpoints/figure5/flat/mlp/seed_6/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f63ddcf99c35b4729a2d044aee254d39abc1b325bd93e9a4b98c52fb883f9d9d
+size 26023117

checkpoints/figure5/flat/mlp/seed_7/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a34dce5248014f878db98eb5519669653486c12fae8579bd9d9fde27392e7e70
+size 25949965

checkpoints/figure5/flat/mlp/seed_8/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24b1f90da82128c6dc3930ad573d10ed5c4da204c46a7129db1dd0541460d887
+size 26071757

checkpoints/figure5/flat/mlp/seed_9/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c50a1098a83475a0fff2857a5200cb840d22029ba12b2f0b5ff1cec4c0dbc25
+size 25962253

checkpoints/figure5/large/adult/seed_0/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da562ef1e80152a6748c929acdbc41e08e3cff260518cb94ca40ebb3b8d45a5b
+size 83924477

checkpoints/figure5/large/adult/seed_1/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:736196ab5a3813a4ce121a3e9efc58c7ae0b4a274384c65859d40702b6dab595
+size 83924477

checkpoints/figure5/large/adult/seed_10/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:599983ec3778408e666e4538a5900c85bd13bd523ef40df3e6b7dc5fba7a440b
+size 83924477

checkpoints/figure5/large/adult/seed_11/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1073fd9068c7b2e0efba66a6e8ff450af2ffeb953a07023921ef422accd731c
+size 83924477

checkpoints/figure5/large/adult/seed_12/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cb93866a8dbd1ce73dbcc3613f365dc2f7ff845a9f4f3c245a20fa96df2c9ab
+size 83924477

checkpoints/figure5/large/adult/seed_13/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16cccdf02090c8617ff2323ce4aa294ccae37a79e9b0bd20fe8a5ad1e7fc4697
+size 83924477

checkpoints/figure5/large/adult/seed_14/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b127ad70620b104891b47ea159d6db6ae0fcb2bfd1c34cc105933138e14a329
+size 83924477

checkpoints/figure5/large/adult/seed_15/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:103e29143abe763cf46e2f5ccb22cd619e59c39d4808653c1eb9a3688fd68caa
+size 83924477

checkpoints/figure5/large/adult/seed_16/final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13ea7cf2c1e68ac8d2ea3beb9abc551bdfba1e88034146bd6759d45e8b65a02e
+size 83924477