File size: 23,057 Bytes
540e67a 6b76d01 d053290 6b76d01 c614c4e 6b76d01 d053290 6b76d01 d053290 6b76d01 3c21768 6b76d01 2bd8497 c614c4e 6b76d01 c614c4e 6b76d01 c614c4e 6b76d01 c614c4e 6b76d01 c614c4e 6b76d01 c614c4e 6b76d01 c614c4e 6b76d01 d053290 6b76d01 3c21768 6b76d01 540e67a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 | {
"title": "Ropedia Xperience-10M Task Suite Project Status",
"version": "2026-06-20",
"decision": "public_sample_pipeline_verified_128_enhancement_qwen3_v6_cosmos_comparison",
"research_positioning": "A research-engineering study that makes one public Xperience-10M sample episode inspectable, defines embodied-AI tasks over synchronized modalities, records baseline behavior, aligns simple/NN baselines to the selected 128-episode split, compares verified Qwen3-Omni and Cosmos3 branch packages as early cross-episode diagnostics, and now records a no-new-episode enhancement pack for pushing the current 128-episode suite harder.",
"scope_boundary": {
"validated_episode_count": 1,
"aligned_frames": 5821,
"sliding_windows": 1161,
"current_feature_dimensions": 8546,
"neural_head_count": 12,
"direction_extension_probe_count": 4,
"audio_featurized": true,
"raw_xperience10m_data_redistributed": false,
"qwen3_omni_32_episode_claim": false,
"qwen3_omni_verified_diagnostic_pilot": true,
"qwen3_omni_selected_episode_counts": {
"train": 96,
"val": 16,
"test": 16
},
"qwen3_omni_exported_window_counts": {
"train": 25629,
"val": 4608,
"test": 4032
},
"qwen3_omni_json_validity_rate": 0.9990079365079365,
"qwen3_omni_validation_aware": true,
"qwen3_omni_json_quality_target_met": true,
"qwen3_omni_lora_adapter_repo": "https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-128ep",
"cosmos3_nano_future_window_compatibility_verified": true,
"cosmos3_nano_future_window_test_predictions": 378,
"cosmos3_super_reasoner_verified": true,
"cosmos3_super_reasoner_test_predictions": 448,
"cosmos3_super_reasoner_json_validity_rate": 0.5111607142857143,
"cosmos3_super_forward_dynamics_lora_verified": true,
"cosmos3_super_forward_dynamics_train_rows": 2848,
"cosmos3_super_forward_dynamics_val_rows": 512,
"cosmos3_super_forward_dynamics_test_rows": 448,
"cosmos3_super_forward_dynamics_test_mse": 3.6853174321087345,
"cosmos3_super_forward_dynamics_adapter_params": 26214400,
"omni_model_comparison_available": true,
"multi_episode_128_aligned_baselines": true,
"multi_episode_128_baseline_window_counts": {
"train": 2848,
"val": 512,
"test": 448
},
"multi_episode_128_baseline_task_count": 20,
"task_method_matrix_method_count": 9,
"task_method_matrix_record_count": 180,
"task_method_matrix_scored_count": 180,
"task_method_matrix_proxy_scored_count": 6,
"qwen3_omni_current_eval_run_id": "xperience10m_qwen3_omni_128ep_multiscale_cap96_v6_rank64_lr5e5_full8gpu_lora_eval_test_full",
"qwen3_omni_current_train_epochs": 2,
"qwen3_omni_action_macro_f1": 0.0028830723979596335,
"qwen3_omni_subtask_accuracy": 0.0037313432835820895,
"qwen3_omni_contact_accuracy": 0.8177083333333334,
"qwen3_omni_object_micro_f1": 0.3064982378331287,
"task_suite_enhancement_128_available": true,
"task_suite_enhancement_128_current_windows": 3808,
"task_suite_enhancement_128_recommended_export": "multiscale_20s10_40s20_80s40",
"task_suite_enhancement_128_estimated_windows": 106095,
"task_count": 20,
"original_public_sample_task_count": 12,
"additional_public_sample_task_count": 8,
"legacy_tasks_13_to_20_result_path": "docs/data/tier2_task_suite.json"
},
"rows": [
{
"area": "Public-sample pipeline",
"status": "verified",
"evidence": [
"results/episode_task_suite/summary_report.json",
"results/episode_task_suite/windows.csv",
"results/episode_task_suite/feature_manifest.json"
],
"readout": "One public Xperience-10M sample episode is converted into 5,821 frames, 1,161 aligned 20-frame windows, and an 8,546-dimensional representation for repeatable task evaluation."
},
{
"area": "Unified 20-task suite",
"status": "verified",
"evidence": [
"TASK_SUITE_20.md",
"docs/data/task_suite_20.json",
"results/episode_task_suite/",
"results/episode_task_suite/tier2_task_suite/"
],
"readout": "All 20 task contracts have committed minimal metrics; tasks 13-20 reuse the same 20-frame windows, 5-frame stride, chronological split, and minimal/neural head pattern. The tier2_task_suite path is historical and now stores tasks 13-20, not a separate public tier."
},
{
"area": "180-result method matrix",
"status": "verified_complete",
"evidence": [
"docs/data/task_method_20_result_matrix.json",
"TASK_METHOD_20_RESULT_MATRIX.md",
"docs/data/task_method_20_gap_audit.json",
"docs/assets/charts/unified_task_model_radar.svg"
],
"readout": "The public comparison matrix now has 9 methods x 20 tasks = 180/180 scored method-task records. Six rows are explicitly marked as compact-proxy scores where the public 128-episode export lacks the direct raw target."
},
{
"area": "Neural heads",
"status": "verified",
"evidence": [
"scripts/neural_task_models.py",
"results/episode_task_suite/neural_mlp/"
],
"readout": "Each task also has a compact PyTorch MLP run over the same feature tensor and chronological split."
},
{
"area": "Audio contribution study",
"status": "verified",
"evidence": [
"scripts/audio_ablation_and_raw_upgrade.py",
"results/audio_ablation/",
"docs/data/audio_ablation_summary.json"
],
"readout": "Audio variants improve the primary metric on 6 of the original task contracts in this single-episode setting."
},
{
"area": "Evaluation protocol",
"status": "verified",
"evidence": [
"EVALUATION_PROTOCOL.md",
"docs/data/evaluation_protocol.json",
"scripts/build_evaluation_protocol.py"
],
"readout": "Windowing, chronological split, per-task metrics, leakage controls, and current limitations are generated from committed metric artifacts."
},
{
"area": "Research takeaways",
"status": "verified",
"evidence": [
"RESEARCH_TAKEAWAYS.md",
"docs/data/research_takeaways.json",
"scripts/build_research_takeaways.py"
],
"readout": "The main result interpretation is generated from committed metrics: chronological class shift, neural gains on dynamics/order/alignment, open retrieval/reconstruction problems, and the need for held-out episodes."
},
{
"area": "Research roadmap",
"status": "current",
"evidence": [
"RESEARCH_ROADMAP.md",
"docs/data/research_roadmap.json"
],
"readout": "The roadmap connects public-sample task development to the final verified Qwen3-Omni diagnostic result, same-split baseline alignment, the no-new-episode 128-suite enhancement pack, action/subtask error analysis, robustness runs, world/policy branches, and the future Xperience-native pretraining goal."
},
{
"area": "128-episode task-suite enhancement pack",
"status": "current_no_new_episode_plan",
"evidence": [
"TASK_SUITE_ENHANCEMENT_128.md",
"docs/data/task_suite_enhancement_128.json",
"results/omni_finetune/task_suite_enhancement_128_v1_20260608/enhancement_plan.json",
"scripts/omni/build_task_suite_enhancement_128.py"
],
"readout": "The current 3,808-window selected split can be stressed without more episodes by exporting denser and multiscale windows. The recommended next export is multiscale_20s10_40s20_80s40, estimated at 106,095 windows from observed frame spans; the pack also defines hierarchical action/subtask targets, raw-feature shard priorities for unsupported tasks, and Qwen/Cosmos follow-up run cards."
},
{
"area": "Foundation-model plan",
"status": "current",
"evidence": [
"FOUNDATION_MODEL_PLAN.md",
"docs/data/foundation_model_plan.json"
],
"readout": "Qwen3-Omni remains the first structured JSON LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package, a verified Cosmos3-Super base-weight Reasoner evaluation, and a verified Cosmos3-Super Forward-Dynamics LoRA over camera-pose proxy targets. The Super LoRA target supports vision-velocity training under action conditioning, not supervised action-token prediction; OpenVLA/openpi/GR00T remain policy candidates after robot-compatible action targets are explicit."
},
{
"area": "Omni model extension contract",
"status": "current",
"evidence": [
"OMNI_MODEL_EXTENSION_CONTRACT.md",
"configs/omni_backbones/",
"scripts/omni/backbone_registry.py",
"scripts/omni/smoke_test_backbone_packaging.py"
],
"readout": "Future Qwen, Cosmos-style, and VLA/policy branches must keep the same episode split discipline, held-out metrics, validation gate, public-safe package contract, and explicit forbidden-artifact policy before reporting results."
},
{
"area": "Xperience Embodied Foundation Model",
"status": "future_goal",
"evidence": [
"XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
],
"readout": "A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model."
},
{
"area": "Official dataset wording",
"status": "verified",
"evidence": [
"XPERIENCE10M_DATASET_CARD_ALIGNMENT.md",
"docs/data/xperience10m_dataset_card_alignment.json"
],
"readout": "Public wording is aligned to the official gated Xperience-10M dataset card, public sample card, and HF API metadata, including modalities, scale, access path, sample license/tooling, and current project coverage."
},
{
"area": "Source alignment",
"status": "verified",
"evidence": [
"SOURCE_ALIGNMENT_AUDIT.md",
"docs/data/source_alignment_audit.json",
"scripts/validate_source_alignment.py"
],
"readout": "Source facts, sample details, API-listing notes, and project coverage are checked across repo docs, website, and HF cards."
},
{
"area": "Website and HF mirrors",
"status": "verified",
"evidence": [
"docs/data/website_integrity.json",
"docs/data/mirror_parity.json",
"docs/data/live_publication_status.json"
],
"readout": "Local website links/assets pass, prepared mirrors match, and public GitHub/HF URLs have been checked after upload."
},
{
"area": "Publication package",
"status": "verified",
"evidence": [
"docs/data/publication_audit.json",
"QUALITY_GATES.md",
"docs/data/quality_gates.json"
],
"readout": "Public bundles are checked for raw-data exclusion, cache exclusion, heavy-archive exclusion, credential-text checks, and current presentation assets."
},
{
"area": "Reproducibility",
"status": "verified_for_public_sample",
"evidence": [
"REPRODUCIBILITY.md",
"docs/data/reproducibility_matrix.json",
"notes/reproducibility_audit.md"
],
"readout": "The public sample workflow has explicit commands, expected outputs, and exact-match reproduction evidence."
},
{
"area": "128-episode aligned baselines",
"status": "verified_companion_result",
"evidence": [
"results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md",
"results/omni_finetune/multi_episode_128_task_baselines/summary_report.json",
"scripts/omni/run_128_task_baselines.py"
],
"readout": "The earlier simple and neural baseline framing is aligned to the selected 96/16/16 episode split used by the Qwen3-Omni pilot. JSON-supported tasks have metadata/text simple and neural MLP metrics; raw-feature-only tasks are explicitly marked unsupported until 128-run sensor feature blocks are available."
},
{
"area": "Current result comparison",
"status": "verified_generated_summary",
"evidence": [
"docs/data/omni_model_comparison.json",
"results/omni_finetune/OMNI_MODEL_COMPARISON.md",
"scripts/omni/build_omni_model_comparison.py"
],
"readout": "The public comparison now has two views: the three result layers and a model-family grouping. The model grouping pairs 1-episode and 128-episode entries for task-head baselines, separates Qwen3-Omni sensor-adapter smoke from 128-episode LoRA diagnostics, separates Cosmos3-Nano future-window compatibility from Cosmos3-Super base-weight Reasoner evaluation, and adds Cosmos3-Super Forward-Dynamics LoRA as a loss-based fine-tuned adapter branch."
},
{
"area": "Qwen3-Omni fine-tuning",
"status": "final_verified_diagnostic_result_json_target_met",
"evidence": [
"docs/data/omni_finetune_verified_result.json",
"docs/data/qwen3_v5_v6_comparison.json",
"results/omni_finetune/QWEN3_V5_V6_COMPARISON_20260614.md",
"results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_multiscale_cap96_v6_rank64_lr5e5_full8gpu_lora_eval_test_full/",
"https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-128ep",
"scripts/omni/package_verified_omni_result.py",
"scripts/omni/audit_verified_omni_package.py",
"scripts/omni/analyze_qwen3_omni_errors.py"
],
"readout": "The selected 96/16/16 episode split now has a current v6 rank64/lr5e-5 public-safe held-out package with 34,269 exported windows, 4,032 test predictions, validation/audit summaries, and a public LoRA adapter repo. JSON validity is 99.90%, meeting the 98% target; transition accuracy is 98.98%, contact accuracy is 81.77%, object micro-F1 is 30.65%, next-action accuracy is 4.31%, and action/subtask metrics remain weak. v6 improves action macro-F1 and contact accuracy versus v5, but v5 remains stronger on JSON validity, subtask, next-action, transition, and object metrics."
},
{
"area": "Cosmos3-Nano future-window branch",
"status": "verified_compatibility_result",
"evidence": [
"configs/omni_backbones/cosmos_world_model.json",
"scripts/omni/export_cosmos3_future_window_dataset.py",
"scripts/omni/eval_cosmos3_future_window_retrieval.py",
"results/omni_finetune/verified_public/xperience10m_cosmos3_nano_128ep_future_window_h5_compat_adapter_eval_test_full/verified_result_summary.json"
],
"readout": "The Cosmos3-Nano branch now has a public-safe verified future-window compatibility package with 3,213 future-window samples, 378 held-out test predictions, future retrieval MRR 0.0221, temporal consistency 0.0952, transition accuracy 0.9683, and contact accuracy 0.7434. It is a compatibility adapter result, not a full Cosmos diffusion-weight fine-tune."
},
{
"area": "Cosmos3-Super Reasoner branch",
"status": "verified_base_weight_result",
"evidence": [
"configs/omni_backbones/cosmos3_super_reasoner.json",
"scripts/omni/eval_cosmos3_super_reasoner.py",
"scripts/omni/run_cosmos3_super_reasoner_eval.sh",
"results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/verified_result_summary.json"
],
"readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
},
{
"area": "Cosmos3-Super action-target contract",
"status": "superseded_by_verified_forward_dynamics_lora",
"evidence": [
"scripts/omni/export_cosmos3_camera_pose_targets.py",
"scripts/omni/pack_cosmos3_super_action_batch.py",
"results/omni_finetune/xperience10m_cosmos3_camera_pose_targets_20260608/target_manifest.json",
"results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
"results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json"
],
"readout": "The selected 128-episode JSONL is augmented with 3,808/3,808 valid camera_pose proxy cosmos_action_target records from SLAM pose deltas. The contract and packer smoke enabled the verified forward-dynamics LoRA run; it supervises noisy vision tokens under camera-pose conditioning and does not supervise preds_action."
},
{
"area": "Cosmos3-Super Forward-Dynamics LoRA",
"status": "verified_fine_tuned_adapter_result",
"evidence": [
"configs/omni_backbones/cosmos3_super_forward_dynamics.json",
"scripts/omni/train_cosmos3_super_forward_dynamics_lora.py",
"scripts/omni/eval_cosmos3_super_forward_dynamics_lora.py",
"results/omni_finetune/verified_public/xperience10m_cosmos3_super_forward_dynamics_lora_128ep_train1epoch_256_attn_full8gpu_20260608_eval_test_full_fsdp/verified_result_summary.json",
"results/omni_finetune/verified_public/xperience10m_cosmos3_super_forward_dynamics_lora_128ep_train1epoch_256_attn_full8gpu_20260608_eval_test_full_fsdp/package_audit.json"
],
"readout": "The first fine-tuned Cosmos3-Super adapter branch is verified as a public-safe package: 8-GPU FSDP LoRA, 26.2M adapter parameters, 2,848 train rows, 512 validation rows, 448 held-out test rows, validation MSE 4.0082, and test MSE 3.6853. The package excludes adapter safetensors; weights are published separately at cy0307/ropedia-cosmos3-super-forward-dynamics-lora-128ep."
},
{
"area": "Raw Xperience-10M redistribution",
"status": "not_included",
"evidence": [
"DATA_NOTICE.md",
"docs/data/publication_audit.json"
],
"readout": "Raw MP4, HDF5, RRD files, private gated data, and full Qwen weights are intentionally excluded."
}
],
"fast_research_route": [
"Read PROJECT_STATUS.md and EVIDENCE_CONTRACT.md to establish what is implemented.",
"Open docs/data/project_packet.json for the machine-readable project path.",
"Inspect RESEARCH_TAKEAWAYS.md and docs/data/research_takeaways.json before interpreting model scores.",
"Inspect RESEARCH_ROADMAP.md and docs/data/research_roadmap.json for the path from public-sample task work to multi-episode modeling.",
"Inspect FOUNDATION_MODEL_PLAN.md and docs/data/foundation_model_plan.json before choosing a backbone branch.",
"Inspect OMNI_MODEL_EXTENSION_CONTRACT.md and run python scripts/omni/backbone_registry.py --validate --json before adding a new Qwen, Cosmos-style, or VLA/policy branch.",
"Inspect XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md for the long-term full-corpus pretraining goal.",
"Inspect TASK_SUITE_20.md, docs/data/task_suite_20.json, docs/data/summary_metrics.json, and results/episode_task_suite/neural_mlp/ to check the unified 20-task outputs.",
"Inspect results/audio_ablation/AUDIO_ABLATION_SUMMARY.md before judging whether audio helps the current task suite.",
"Inspect EVALUATION_PROTOCOL.md before judging task metrics or leakage controls.",
"Inspect SOURCE_ALIGNMENT_AUDIT.md before judging source-card consistency across public surfaces.",
"Inspect XPERIENCE10M_DATASET_CARD_ALIGNMENT.md before judging dataset wording.",
"Inspect docs/data/task_method_20_result_matrix.json and TASK_METHOD_20_RESULT_MATRIX.md before comparing the 180 scored method-task records.",
"Inspect results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md and results/omni_finetune/a100_128_metadata_task_baselines_20260616_v2/ before comparing simple/NN baselines to the selected 128-episode setup.",
"Inspect TASK_SUITE_ENHANCEMENT_128.md and docs/data/task_suite_enhancement_128.json before deciding whether more episodes are needed; the current recommended no-new-episode export is multiscale_20s10_40s20_80s40.",
"Inspect docs/data/omni_model_comparison.json before comparing the current three result versions or the model-family 1-episode versus 128-episode groupings.",
"Inspect docs/data/omni_finetune_verified_result.json before judging the Qwen3-Omni diagnostic pilot."
],
"current_reading_notes": [
"The latest Qwen3-Omni v6 diagnostic branch is verified and meets the strict-JSON target, but action/subtask held-out quality is still weak: JSON validity is 99.90%, action macro-F1 is 0.0029, and subtask accuracy is 0.0037. v5 remains the pinned prior release row because it is still stronger on several metrics.",
"Use TASK_SUITE_ENHANCEMENT_128.md and docs/data/task_suite_enhancement_128.json to push the current 128-episode suite without more raw episodes through multiscale_20s10_40s20_80s40, hierarchical labels, label-normalized scoring, and raw-feature shard export.",
"Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
"The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
"The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, Cosmos3-Super Reasoner is verified as a base-weight evaluation, and Cosmos3-Super Forward-Dynamics LoRA is verified as the first fine-tuned Super adapter branch. Cosmos3-Super adapter weights belong in cy0307/ropedia-cosmos3-super-forward-dynamics-lora-128ep; verified_public packages exclude safetensors.",
"The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
"Audio is one of the synchronized source modalities in the current task representation.",
"The audio ablation report compares audio/no-audio variants across the original task contracts in results/audio_ablation/.",
"Foundation-model selection is explicit: Qwen3-Omni is the structured JSON baseline, Cosmos 3 is the world-model branch with Nano compatibility and Super forward-dynamics LoRA results, and policy models such as OpenVLA/openpi/GR00T wait for robot-compatible action-target conversion.",
"Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
"The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
]
}
|