Add real-time RoboCasa environment graph extraction pipeline

docs/SCHEMA.md

@@ -1,659 +1,67 @@
-# GWAM_Data Graph Schema v1
+# GWAM_Data v1.2 sparse-v2 schema
 
-This schema defines the first graph-augmented RoboCasa export for Graph-WAM / DreamZero-style joint vision+action world modeling.
-
-V1 intentionally excludes depth. Depth is a v2 extension.
-
-## Core principle
-
-Each timestep has one global scene graph:
-
-```text
-G_t = (V_t, E_state_t, E_prior, globals_t)
-```
-
-Camera views are evidence attached to global nodes, not separate graphs.
-
-```text
-robot0_agentview_right  -> node view evidence
-robot0_agentview_left   -> node view evidence
-robot0_eye_in_hand      -> node view evidence
-```
-
-## Cameras
-
-Canonical names:
-
-```yaml
-side_right: robot0_agentview_right
-side_left: robot0_agentview_left
-wrist: robot0_eye_in_hand
-primary_side: side_right
-```
-
-All three cameras are exported in v1. `side_right` is canonical only for single-side experiments and report ordering.
-
-## Node definition
-
-A node is a task-relevant component, object, fixture, surface, articulated part, or optional robot aggregate.
-
-Recommended families:
-
-```text
-object
-distractor
-surface
-receptacle
-fixture
-articulated_part
-handle
-appliance_part
-robot_optional
-```
-
-Do not create one node per MuJoCo geom. Group geoms/bodies into task-relevant components using a deterministic `node_grouping_<task>.yaml` file.
-
-## Node record
-
-Each node in `graph_static.json` contains:
-
-```json
-{
-  "node_id": "obj_000",
-  "family": "object",
-  "category_text": "mug",
-  "instance_name": "mug_1",
-  "body_names": ["mug_1_main"],
-  "body_ids": [123],
-  "geom_ids": [456, 457],
-  "joint_ids": [],
-  "movability": {
-    "free_movable": true,
-    "articulated": false,
-    "static": false
-  },
-  "type_encoding": {
-    "encoder": "openai/clip-vit-base-patch32-text",
-    "raw_dim": 512,
-    "stored_dim": 32,
-    "projection": "pca_or_random_projection_frozen_v1",
-    "vector_ref": "node_type_clip32.npy:0"
-  },
-  "source": "sim_metadata",
-  "confidence": 1.0
-}
-```
-
-## Per-frame node state
-
-Each node in each `frames.jsonl` timestep contains:
-
-```json
-{
-  "node_id": "obj_000",
-  "position_world": [0.0, 0.0, 0.0],
-  "orientation_world_xyzw": [0.0, 0.0, 0.0, 1.0],
-  "linear_velocity": [0.0, 0.0, 0.0],
-  "angular_velocity": [0.0, 0.0, 0.0],
-  "qpos": null,
-  "qvel": null,
-  "visibility": {
-    "side_right": {
-      "visible": true,
-      "mask_ref": "seg/side_right/frame_000123.npz:obj_000",
-      "bbox_xyxy": [12, 34, 90, 110],
-      "area_px": 1500,
-      "centroid_xy": [50.2, 70.1]
-    },
-    "side_left": {"visible": false},
-    "wrist": {"visible": true}
-  }
-}
-```
-
-No depth fields are included in v1.
-
-## Node feature tensor
-
-`graph_tensors.npz` stores:
-
-```text
-node_feats: float32[T, N_max, D_node]
-node_mask: bool[N_max]
-```
-
-Recommended v1 node feature block:
+This file mirrors the root `SCHEMA.md`. The current v1.2 sparse-v2 export uses:
 
 ```text
-pose_position_world                 3
-pose_orientation_6d                 6
-linear_velocity                     3
-angular_velocity                    3
-qpos_scalar_or_zero                 1
-qvel_scalar_or_zero                 1
-movability_free_movable             1
-movability_articulated              1
-family_onehot                       8-10
-clip_type_text_projected           32
-visibility_side_right               1
-visibility_side_left                1
-visibility_wrist                    1
-mask_area_side_right_norm           1
-mask_area_side_left_norm            1
-mask_area_wrist_norm                1
-bbox_center_side_right_xy_norm      2
-bbox_center_side_left_xy_norm       2
-bbox_center_wrist_xy_norm           2
-bbox_size_side_right_wh_norm        2
-bbox_size_side_left_wh_norm         2
-bbox_size_wrist_wh_norm             2
+graph/node_state.npz: node_state float32[T, 256, 32]
+graph/view_evidence.npz
+graph/dynamic_edges.jsonl with contact + visibility_change
+graph/visible_masks_rle.jsonl.gz with row-major binary RLE v2_vectorized
+audit/phase2_summary.json after Phase 2
+sparse visible-only Phase 2 visual features
 ```
 
-Default Fable-reviewed tensor view for full v1 uses D_node = 317:
+Old dense v1/v1.1 concepts are historical only and are not emitted by v1.2 sparse-v2:
 
 ```text
-[0:256]   visual_256 from masked visual encoder / SAM2-on-oracle-mask using canonical side mask
-[256:259] position_world xyz
-[259:263] orientation quaternion xyzw
-[263:266] linear velocity
-[266:269] angular velocity
-[269:270] qpos normalized or zero
-[270:271] qvel normalized or zero
-[271:272] binary appliance/control state or zero
-[272:274] movability [free_movable, articulated]
-[274:275] robot contact flag
-[275:277] visibility [side_right, wrist]
-[277:285] family one-hot, 8 slots
-[285:317] type_clip_32
-```
-
-A compact loader may construct a smaller feature vector, but the published default tensor view is 317D and the slicing map must be saved in `metadata/tensor_view_v1.json`.
-
-### CLIP component type encoding
-
-Store three things:
-
-1. raw `category_text`, e.g. `"drawer handle"`, `"cabinet"`, `"mug"`;
-2. CLIP metadata:
-
-```json
-{
-  "encoder": "openai/clip-vit-base-patch32-text",
-  "raw_dim": 512,
-  "normalization": "l2",
-  "prompt_template": "a photo of a {category_text}",
-  "stored_dim": 32,
-  "projection": "frozen_pca_or_random_projection_v1"
-}
+317-D / 323-D node_feats
+dense rel_state or rel_prior tensors
+distractor as family index 1
+events.jsonl state_on/state_off / grasp / release vocab
+dense visual_views[T,N,3,256]
+min_visible_px=10 or RLE v1
 ```
 
-3. compressed vector in tensors and optionally raw vector sidecar:
+Current family ids:
 
 ```text
-node_type_clip32.npy
-node_type_clip512_optional.npy
+0 object, including distractors
+1 reserved / forbidden old distractor id
+2 surface
+3 receptacle
+4 fixture
+5 articulated_part
+6 handle
+7 control, defined but unpopulated
+8 region, defined but unpopulated
+9 robot
 ```
 
-Recommendation: use frozen CLIP text embeddings, not trainable text encodings. The 32D projected vector is the default model-facing feature. The full 512D CLIP vector is stored once per type in `encodings/type_encodings_v1.npz` or equivalent; raw category text and encoder metadata are authoritative.
-
-## Edge tiers
-
-### State-tier edges
+Loaders must read `node_state_schema`, gate on the active channel before using padded rows, and treat instruction/goal/object-role metadata as supervision-only unless an experiment explicitly allows it.
 
-State-tier edges represent current physical/geometric state at timestep `t`.
+See root `README.md` and root `SCHEMA.md` for the full normative contract.
 
-Allowed families:
 
-```text
-support_contact
-robot_contact
-held_by_gripper
-candidate_spatial
-inside_current
-articulation_state
-```
+## Live real-time environment graph compatibility
 
-Sources:
+The on-disk release stores full episodes, while the real-time extractor returns one current simulator frame at a time. They use the same schema vocabulary:
 
 ```text
-sim_contacts
-sim_qpos
-sim_pose_geometry
-robot_state
+on-disk episode:        node_state float32[T,256,32]
+live current frame:     node_state float32[256,32]
+live K-frame buffer:    node_state float32[K,256,32]
 ```
 
-Never source state-tier edges from a reasoning model.
-
-### Prior-tier edges
-
-Prior-tier edges represent static semantic, vocabulary, or structural relations.
-
-Allowed families:
-
-```text
-articulation_part
-kinematic_part_of
-precondition_affordance
-category_compatibility
-```
-
-Sources:
-
-```text
-sim_kinematic_tree
-sim_metadata
-human_reviewed_rule_table
-```
-
-Reserved future source values:
-
-```text
-llm_prior
-vlm_caption
-manual_annotation
-```
-
-In sim v1, do not require reasoning models. If an LLM drafts the rule table, mark the committed source as `human_reviewed_rule_table` after review.
-
-## Edge record
-
-```json
-{
-  "src": "obj_000",
-  "dst": "fixture_001",
-  "tier": "state",
-  "family": "support_contact",
-  "relation_type": "resting_on",
-  "active": true,
-  "directed": true,
-  "source": "sim_contacts",
-  "confidence": 1.0,
-  "attrs": {
-    "num_contacts": 4,
-    "min_contact_dist": -0.002,
-    "relative_position": [0.1, 0.0, 0.2]
-  }
-}
-```
-
-## Edge tensors
-
-```text
-rel_state: float32[T, N_max, N_max, C_state]
-rel_prior: float32[N_max, N_max, C_prior]
-```
-
-Suggested `C_state` channels:
-
-```text
-active_contact
-supporting
-supported_by
-robot_touching
-held_by_gripper
-inside_current
-near
-above
-articulation_open_scalar
-relative_position_xyz             3
-relative_distance                 1
-```
-
-Suggested `C_prior` channels:
-
-```text
-part_of
-controls_articulation
-can_support
-can_contain
-can_be_placed_in
-same_fixture_group
-category_compatible
-```
-
-Goal/instruction-derived edges must not enter `rel_state` or `rel_prior`. Goal grounding lives in globals only.
-
-## Globals
-
-Each episode has:
-
-```json
-{
-  "dataset_version": "robocasa_target_atomic_seen_graph_v1",
-  "task": "OpenDrawer",
-  "split": "target_atomic_seen",
-  "source": "human",
-  "language_instruction": "Open the drawer.",
-  "control_freq": 20,
-  "primary_side_camera": "robot0_agentview_right",
-  "exported_cameras": ["robot0_agentview_right", "robot0_agentview_left", "robot0_eye_in_hand"],
-  "graph_source": "oracle_sim",
-  "depth_included": false,
-  "goal_grounding": {
-    "supervision_only": true,
-    "target_node_ids": [],
-    "goal_relation_set": []
-  }
-}
-```
-
-## Events / ΔG
-
-`events.jsonl` records relation changes and graph events:
-
-```json
-{"frame": 123, "kind": "contact_make", "src": "obj_000", "dst": "surface_000", "from": false, "to": true}
-```
-
-Event kinds:
-
-```text
-contact_make
-contact_break
-support_change
-grasp
-release
-articulation_open_threshold
-articulation_close_threshold
-inside_make
-inside_break
-visibility_appeared
-visibility_disappeared
-goal_satisfied_eval_only
-```
-
-## Leakage rules
-
-- No instruction-derived target edge in `rel_state` or `rel_prior`.
-- No reasoning-model output in current physical state fields.
-- Prior edges must be broadcast by category/structure, not only attached to the target pair.
-- Graph inventory must be deterministic from sim metadata and task configs, not filtered by language relevance.
-- Store `source` for every node/edge/global field.
-
-
-## Published layout vs extraction working layout
-
-The extraction workspace may use per-frame PNG/NPZ files because they are easy to inspect. The Hugging Face published data should avoid millions of files. Preferred published layout after data exists:
+The live extractor in `realtime/gwam_realtime_env_graph.py` / `extractors/gwam_realtime_env_graph.py` uses the same:
 
 ```text
-data/v1_no_depth/
-  meta/info.json
-  meta/tasks.jsonl
-  meta/episodes.jsonl
-  videos/chunk-000/observation.images.robot0_agentview_right/episode_000000.mp4
-  videos/chunk-000/observation.images.robot0_agentview_left/episode_000000.mp4
-  videos/chunk-000/observation.images.robot0_eye_in_hand/episode_000000.mp4
-  graphs/chunk-000/episode_000000.graph.jsonl.zst
-  masks/chunk-000/episode_000000.masks.npz
-  masks/chunk-000/episode_000000.mask_index.jsonl.zst
-  tensors/chunk-000/episode_000000.graph_tensors.npz
-  audit/chunk-000/episode_000000_summary.json
+node_state_schema
+family ids
+active-mask policy
+static prior edge labels: part_of, controls
+dynamic edge labels: contact, visibility_change
+view_evidence fields and normalized [0,1] units
+row-major RLE mask codec when masks are requested
 ```
 
-This LeRobot-like chunked layout is the publication target. Per-episode directories in the extraction plan are an intermediate/debugging layout.
-
-
-# Schema v1.1 normative update
-
-
-
-Fable 5 and Codex both reviewed the six design questions about multi-view graphs, velocity, feature balance, source availability, task-family coverage, and vocabulary edges. The v1.1 clarification is now the normative design.
-
-### 1. One global graph, not one graph per camera
-
-Use one authoritative global graph per timestep:
-
-```text
-G_t = (V_t, E_state_t, E_prior, globals_t)
-```
-
-Do not create separate physical graphs for `robot0_agentview_right`, `robot0_agentview_left`, and `robot0_eye_in_hand`.
-
-Reason:
-
-- Object identity is global. A mug visible in the right camera and wrist camera is the same MuJoCo body/node.
-- Physical relations are global. Contact, support, containment, qpos, held state, and robot interaction do not become false because a camera cannot see one endpoint.
-- Per-camera graphs would confuse occlusion with object deletion and make physical edges view-dependent.
-- RoboCasa/MuJoCo already gives free cross-view association through `segmentation objid -> geom -> body -> node`.
-
-But the user’s concern is correct: camera evidence is different. Therefore v1.1 stores **one global graph with per-view node evidence and per-view visual features**.
-
-Normative multi-view tensors:
-
-```text
-visual_views   float16[T, N_max, 3, 256]   # per-view masked RGB/SAM2-style visual features
-view_visible   bool[T, N_max, 3]
-view_centroid  float16[T, N_max, 3, 2]
-view_area      float16[T, N_max, 3, 1]
-```
-
-View order:
-
-```text
-0 = robot0_agentview_right
-1 = robot0_agentview_left
-2 = robot0_eye_in_hand
-```
-
-The canonical side visual feature remains in the flat compatibility node tensor, but the complete visual evidence is `visual_views`.
-
-### 2. Velocity definition
-
-Velocity means node-level physical velocity from the simulator, source-tagged.
-
-Recommended source in RoboCasa/MuJoCo:
-
-```text
-mujoco.mj_objectVelocity(model, data, mjOBJ_BODY, body_id, res, flg_local=0)
-```
-
-This gives 6D world-frame body velocity. Store as linear and angular world velocity after confirming convention in the exporter.
-
-Do not use raw free-joint `qvel` as object angular velocity without care: MuJoCo free-joint angular velocity conventions can be body-local. Use `mj_objectVelocity` for body-level velocity and store raw `qvel` only for joint/articulation state.
-
-Per family:
-
-| Family | Body velocity | Joint velocity |
-|---|---|---|
-| `object` / movable item | root body world velocity | zero unless jointed |
-| `articulated_part` | part body world velocity | scalar qvel is primary articulation signal |
-| `control` | body velocity if meaningful | scalar qvel / control state if jointed |
-| `surface` / `receptacle` / `fixture` | zero, with `velocity_source=static_fixture` | zero unless fixture state joint exists |
-| optional robot node | TCP/eef velocity from robot state or MuJoCo body velocity | robot proprio stream remains primary |
-
-Add QA:
-
-```text
-finite_difference(position_world) ~= stored_linear_velocity
-```
-
-on contact-free segments, and assert static fixtures have zero velocity.
-
-### 3. Feature balance: 256D visual block should not dominate by raw concat
-
-The dataset stores typed feature blocks. It does not recommend feeding a raw 300D+ concatenation directly into a GNN.
-
-Downstream models should use per-block projections:
-
-```text
-static_proj = MLP(node_static_feats)
-dynamic_proj = MLP(node_dynamic_feats)
-view_proj = PerViewEncoder(view_geom_feats, visual_views, view_visible)
-edge_proj = EdgeEncoder(rel_state, rel_prior)
-```
-
-Then fuse into common `d_model` tokens.
-
-Export rules:
-
-- L2-normalize masked visual embeddings.
-- Store per-block normalization stats.
-- Keep visual features physically separate from symbolic state features.
-- Publish a `symbolic_only_v1_1` tensor view for ablation.
-- Treat the old flat `node_feats` as a compatibility view, not the only model input.
-
-### 4. Final v1.1 tensor layout
-
-Published graph tensor file:
-
-```text
-graph_tensors.npz
-  node_feats        float32[T, N_max, 323]       # flat compatibility typed-block view
-  node_mask         bool[N_max]
-  category_ids      int16[N_max]
-
-  visual_views      float16[T, N_max, 3, 256]
-  view_visible      bool[T, N_max, 3]
-  view_centroid     float16[T, N_max, 3, 2]
-  view_area         float16[T, N_max, 3, 1]
-
-  rel_state         float32[T, N_max, N_max, 8]
-  rel_prior         float32[N_max, N_max, 5]
-  event_masks       uint8[T, 13]
-```
-
-`node_feats` v1.1 slicing map, D = 323:
-
-```text
-[0:256]    visual_256        canonical-side visual feature = visual_views[:, :, 0]
-[256:259]  pos_world_3
-[259:263]  quat_xyzw_4
-[263:266]  linvel_world_3
-[266:269]  angvel_world_3
-[269:270]  qpos_norm_1
-[270:271]  qvel_norm_1
-[271:272]  binary_state_1
-[272:274]  movability_2      [free_movable, articulated]
-[274:275]  robot_contact_1
-[275:276]  held_1
-[276:279]  visibility_3      [side_right, side_left, wrist]
-[279:291]  family_12
-[291:323]  type_clip_32
-```
-
-### 5. Updated node families
-
-Use 12 frozen family slots. Family is the relational role, not the full semantic class. Semantic class is carried by `category_text`, `category_id`, and CLIP type encoding.
-
-```text
-0  object              free movable item: mug, can, food
-1  distractor          free movable non-task item; metadata-derived, not language-derived
-2  surface             static support plane: counter, stove top
-3  receptacle          containment volume: cabinet/drawer/fridge/sink basin
-4  fixture             appliance/furniture body: stove, microwave, kettle, blender, fridge
-5  articulated_part    door, drawer, lid, rack, mixer head
-6  handle              grasp point rigidly attached to articulated part
-7  control             knob, button, switch, faucet lever; actuates fixture state
-8  region              named zone: burner zone, rack slot, navigation region; v1 optional/reserved
-9  robot               optional aggregate robot/TCP node for ablations
-10 reserved
-11 reserved
-```
-
-`appliance_part` is removed because it is ambiguous. Use `articulated_part`, `handle`, or `control` instead.
-
-### 6. Updated relation tensors
-
-State relation tensor:
-
-```text
-rel_state: float32[T, N_max, N_max, 8]
-channels:
-  contact
-  support
-  near
-  inside
-  rel_dx
-  rel_dy
-  rel_dz
-  dist_norm
-```
-
-Prior relation tensor:
-
-```text
-rel_prior: float32[N_max, N_max, 5]
-channels:
-  part_of
-  controls
-  same_fixture_group
-  can_contain
-  can_support
-```
-
-Direction conventions:
-
-```text
-support[i,j] = 1       means node i rests on/supports-contact with node j as support destination
-inside[i,j] = 1        means node i is inside node j
-part_of[i,j] = 1       means node i is part of node j
-controls[i,j] = 1      means node i actuates/controls node j
-can_contain[i,j] = 1   means node i can contain node j
-can_support[i,j] = 1   means node i can support node j
-```
-
-Robot contact and held state are node flags by default. Robot-object edges are created only in a robot-node ablation.
-
-### 7. Vocabulary information
-
-Node vocabulary is included:
-
-```text
-category_text
-category_id
-type_clip_32
-full CLIP 512D type table
-```
-
-Edge vocabulary is primarily discrete and rule-table based:
-
-```text
-relation channel IDs
-human-reviewed affordance matrix
-sim kinematic tree relations
-```
-
-Optional metadata may include CLIP embeddings of relation names, but relation CLIP vectors are not part of the default relation tensor.
-
-Reasoning/VLM/LLM policy:
-
-- No reasoning-model dependency in sim v1.
-- LLM may draft rule tables, but committed source is `human_reviewed_rule_table`.
-- `category_text` comes only from fixed rename tables keyed by asset/sim metadata; no per-episode language edits.
-- No reasoning output is allowed for contact, support, held state, qpos, current inside state, or physical relations.
-
-### 8. Feature source table
-
-| Feature | Source in v1 | Needs CLIP/VLM/LLM? |
-|---|---|---|
-| node inventory | sim metadata + deterministic grouping rules | no |
-| body/geom/joint IDs | MuJoCo model | no |
-| family | deterministic role rules | no |
-| category_text/category_id | fixed rename table from sim metadata | no |
-| type_clip_32 | frozen CLIP text encoder | CLIP yes, VLM/LLM no |
-| RGB | replay render or stored video | no |
-| masks/segmentation | MuJoCo segmentation render | no |
-| bbox/centroid/area | masks | no |
-| visual_views 256D | masked RGB visual encoder pooled by oracle mask | visual encoder yes; not reasoning |
-| pose/orientation | sim state xpose/xquat | no |
-| velocity | `mj_objectVelocity`; FD only QA/fallback | no |
-| qpos/qvel | sim qpos/qvel via joint addresses | no |
-| binary state | qpos/state thresholds | no |
-| contact/support | sim contacts + deterministic filter/rule | no |
-| inside/near/relative geometry | sim pose/AABB geometry | no |
-| part_of/controls | kinematic tree/fixture metadata | no |
-| can_contain/can_support | human-reviewed rule table | no LLM dependency |
-| target labels | task metadata | supervision/eval only |
-| depth-derived geometry | not in v1 | not available |
-
-### 9. New sanity checks
-
-- Swapping instruction text must not change node features, edge features, node inventory, or relation tensors.
-- A node hidden in one camera but visible in another remains one global node.
-- `visual_views[t,n,v]` is valid only when `view_visible[t,n,v]` is true.
-- `visual_views[:, :, 0]` must match `node_feats[:, :, 0:256]` for the compatibility block.
-- Every category ID resolves in the type table.
-- Static fixtures have zero velocity.
-- FD velocity approximately agrees with stored velocity on contact-free segments.
-- Affordance channels must reproduce exactly from the published matrix gather.
-- Event vocabulary must match across README, SCHEMA, relation vocab, and exporter output.
+Real-time extraction is for users with their own RoboCasa/robosuite environment and assets. It constructs a current-observed graph from the current simulator state and segmentation renders. It does not compute the offline Phase-2 SAM2/CLIP `visual_features_sparse.npz` by default. Future simulator states, future robot nodes, future contacts, goal-derived target-node ids, object role labels, and instruction-derived target edges are not default-safe evaluation inputs; use them only in explicitly labeled oracle or teacher-forced modes.