Publish milo_benchmark v1.0: 25-task/5-scene/3-tier dataset with real 3-planner baseline

README.md

@@ -0,0 +1,220 @@
+---
+pretty_name: MILO Benchmark
+language:
+  - en
+license: mit
+task_categories:
+  - robotics
+tags:
+  - embodied-ai
+  - ai2thor
+  - task-planning
+  - robotics
+  - synthetic
+size_categories:
+  - n<1K
+---
+
+# MILO Benchmark v1.0
+
+A small, versioned dataset of `(scene, instruction, ground-truth task
+spec)` triples for evaluating embodied task planning in AI2-THOR,
+built for the [MILO vision-language-robotics
+project](https://github.com/). Every task pairs a natural-language
+instruction with a structured goal/object/target spec and a
+machine-checkable success predicate, across 5 real AI2-THOR scenes
+spanning all 4 iTHOR room types.
+
+**This is a synthetic, AI2-THOR-derived dataset, not a human-collected
+one.** Every instruction was authored by a human against a live scan
+of each scene's real object inventory (`get_metadata()`), not
+generated by an LLM and not crowd-sourced — see "Collection
+methodology" below for exactly how, so nobody mistakes this for
+naturalistic human instruction data.
+
+## What's in it
+
+25 tasks across 5 scenes:
+
+| Scene | Room type | Tasks |
+|---|---|---|
+| `FloorPlan1` | kitchen | 5 |
+| `FloorPlan5` | kitchen | 5 |
+| `FloorPlan201` | living room | 5 |
+| `FloorPlan301` | bedroom | 5 |
+| `FloorPlan401` | bathroom | 5 |
+
+Three difficulty tiers, 5 tasks/scene (2 tier1, 2 tier2, 1 tier3):
+
+| Tier | What it exercises | Example |
+|---|---|---|
+| `tier1_locate` | Single-step object resolution (no manipulation). | "Find the mug." |
+| `tier2_pickup` | Navigate + pick up a named object. | "Pick up the apple." |
+| `tier3_store` | Pick up an object, navigate to a receptacle, open it if needed, place the object, close it if it was opened. | "Put the bread away in the fridge." |
+
+Each row (see `tasks.json`):
+
+```json
+{
+  "task_id": "milo-v1-fp1-t3a",
+  "scene": "FloorPlan1",
+  "room_type": "kitchen",
+  "difficulty_tier": "tier3_store",
+  "instruction": "Put the bread away in the fridge.",
+  "goal": "store",
+  "object": "bread",
+  "target": "fridge",
+  "notes": "fridge is a confirmed openable container in this scene."
+}
+```
+
+`goal`/`object`/`target` map directly onto this project's
+`schemas.task.SingleTask` (`goal` is a canonical verb like `find`/
+`pick_up`/`store`; a planner unrelated to MILO can just as easily
+treat them as generic action/argument fields). `notes` is
+non-scoring, human-readable context — for a handful of tasks it
+documents a **known, real limitation** the task deliberately keeps
+rather than hides (see below).
+
+## Success predicate
+
+A task is scored `goal_success = True` iff its goal condition holds
+against **live** AI2-THOR object state after execution (not just "did
+every planned action dispatch without an error" — those are two
+different claims; both are reported separately by the reference
+runner):
+
+| Goal | Live predicate |
+|---|---|
+| `find`/`locate`/... | An object of the named type exists in the scene. (See "Known limitations" — this is necessarily weaker than a real perception check.) |
+| `pick_up`/`fetch`/`deliver` | The named object's live `isPickedUp` is `True`. |
+| `store`/`place`/`put_away` | The named object is not held, and its live `parentReceptacles` includes the target's `objectId`. |
+| `open` / `close` | The named object's live `isOpen` is `True` / `False`. |
+
+Reference implementation: `backend/planning_evaluation/live_state.py`'s
+`check_goal_live()` in the [MILO
+repository](../../../..) (this dataset's origin repo).
+
+## Baselines (v1.0, real runs, all three planners)
+
+| Planner | Goal success | tier1_locate | tier2_pickup | tier3_store | Notes |
+|---|---|---|---|---|---|
+| `rule_based` | 24/25 (96%) | 10/10 | 10/10 | 4/5 | Deterministic, no LLM. One failure is a real AI2-THOR placement/geometry limit, not a planner defect. |
+| `behavior_tree` | 24/25 (96%) | 10/10 | 10/10 | 4/5 | Same task/plan-step outcomes as `rule_based` (shares its goal-handler templates); same single failure. |
+| `react` (`qwen2.5:7b`, Q4_K_M, via Ollama, local) | 20/25 (80%) | 10/10 | 10/10 | 0/5 | `goal_success`/`execution_success`/`plan_success` agree on every episode — no predicate artifact. All 5 failures are genuine multi-step reasoning failures (the model proposes an action before its precondition chain is satisfied, e.g. `pickup` before navigating close enough), not infrastructure. Run on an RTX 4050 Laptop GPU (6GB VRAM, 82%/18% GPU/CPU split), zero rate-limit retries needed (fully local, no quota). |
+
+`react` was also attempted against Gemini's free tier
+(`gemini-flash-latest`) first; that attempt is **not** a valid
+baseline and is excluded from the table above — the free tier's daily
+quota (20 requests/day) was exhausted after 2 of 25 episodes, and a
+raw success-rate computed from that run would have been actively
+misleading (most of its apparent "successes" were `tier1_locate`
+episodes where the LLM call had already failed outright — the
+predicate can't distinguish "the agent found it" from "the object was
+already sitting in the scene regardless of what the agent did"). See
+the origin repository's `experiments/reports/
+phase_e_milo_benchmark_report.md` (Addendum 2 for the Gemini attempt
+and why it doesn't count, Addendum 3 for the `qwen2.5:7b` run this
+table reports) for full methodology, exact commands, and reproduction
+steps.
+
+**Reproducing the `react` row**: any OpenAI-API-compatible local
+server works (Ollama, vLLM, ...) — set `LANGUAGE_LLM_PROVIDER=qwen`,
+`LANGUAGE_LLM_MODEL=qwen2.5:7b` (or your chosen model/quantization),
+`LANGUAGE_LLM_BASE_URL` to your server's `/v1` endpoint, and
+`QWEN_API_KEY` to any placeholder value if your server doesn't enforce
+auth, then run
+`RUN_SIMULATOR_TESTS=true python -m planning_evaluation.run_benchmark`
+from the origin repository's `backend/` directory.
+
+## Difficulty tiers and why they were chosen this way
+
+Tier boundaries were chosen to exercise structurally different code
+paths in MILO's own rule-based planner (object resolution only, vs.
+navigate+pickup, vs. the full open/place/close container logic) — not
+an arbitrary linguistic complexity scale. This is a deliberate design
+choice: the project's own bug history showed these three code paths
+fail independently (a closed-receptacle bug and a non-openable-target
+bug both lived specifically in the `tier3_store` code path, never in
+`tier1_locate`/`tier2_pickup`) — see "Known limitations" below for the
+two specific, currently-still-open bugs this dataset intentionally
+keeps as ground truth.
+
+## Known limitations — kept deliberately, not hidden
+
+Two `tier3_store` tasks are **known, currently reproducible failures**
+against the reference planner, kept in v1.0 on purpose as honest
+negative examples rather than removed to inflate a headline number:
+
+- `milo-v1-fp301-t3a` ("Put the book away in the drawer.") fails at
+  execution due to a real AI2-THOR physics/geometry limit — the
+  drawer opens correctly, but AI2-THOR cannot find room for this
+  particular book inside this particular drawer's real interior
+  volume. Not a planner defect.
+- `milo-v1-fp401-t3a` ("Put the spray bottle away on the shelf.")
+  fails because the reference rule-based planner tries to `open` the
+  shelf before placing — a shelf is a valid receptacle but is not
+  openable, and the planner's `_deposit()` logic doesn't yet check
+  `openable` before deciding to open. A genuine, reproducible planner
+  bug, tracked in the origin repo's roadmap.
+
+A different, honest limitation of the success predicate itself:
+`tier1_locate`'s live check (existence of an object of the named
+type) cannot verify the agent actually *perceived* the object — only
+that the plan named a real object. A perception-grounded check would
+need a vision pipeline wired into the scoring harness; this dataset's
+reference runner does not do that yet (see the origin repo's Phase C
+vision-grounding work, which is not yet connected to this benchmark).
+
+## Collection methodology
+
+1. For each of the 5 candidate scenes, a live AI2-THOR
+   `Controller.step()`/`last_event.metadata` scan was taken to list
+   every real object's `objectType`, `pickupable`, `receptacle`, and
+   `openable` flags.
+2. Task objects/targets were chosen only from that confirmed live
+   list — never guessed from AI2-THOR documentation or an LLM's
+   assumption about what "should" be in a kitchen/bedroom/bathroom.
+3. Instructions were hand-written in natural language to match each
+   task spec (not generated by an LLM, not templated beyond the
+   tier's basic sentence shape).
+4. `tier3_store` targets were chosen to include both confirmed real
+   containers (fridge/cabinet/drawer) and one confirmed non-container
+   receptacle (shelf) deliberately, once a first sweep run (this
+   project's own "Phase D" floor-plan generalization sweep) surfaced
+   the non-openable-target bug — see "Known limitations" above.
+
+This is the same authoring discipline the origin repository already
+used for its `FloorPlan1`-only real-AI2-THOR task sets
+(`real_scenarios.py`), extended across scenes.
+
+## What this dataset does not cover
+
+- Only 5 of iTHOR's ~120 scenes (one per room type, plus a second
+  kitchen) — not a claim of full scene coverage.
+- Only single-object, single-goal tasks — no multi-object, multi-step,
+  or conditional instructions.
+- No adversarial/ambiguous instructions (Language-layer clarification
+  behavior is out of scope here).
+- English only.
+
+## Versioning
+
+`v1.0` is frozen — task IDs, scenes, and success predicates in this
+version will not change. Future versions extend rather than mutate
+(e.g. `v1.1` adding scenes/tasks would live in a sibling `v1.1/`
+directory with its own `tasks.json`), so a score reported against
+`v1.0` stays reproducible indefinitely.
+
+## License
+
+MIT, matching the origin repository. AI2-THOR scene assets themselves
+are licensed separately by their own maintainers (Allen Institute for
+AI) — this dataset contains no scene assets, only task
+specifications/instructions referencing public AI2-THOR scene IDs.
+
+## Citation
+
+This is a research-adjacent project artifact, not a peer-reviewed
+publication. If referencing it, cite the origin repository and this
+dataset version (`milo_benchmark v1.0`).

tasks.json

@@ -0,0 +1,43 @@
+{
+  "dataset": "milo_benchmark",
+  "version": "1.0",
+  "description": "Versioned (scene, instruction, ground-truth task spec) triples for evaluating embodied task planning in AI2-THOR, generated for the MILO vision-language-robotics project.",
+  "generated_by": "backend/planning_evaluation, following the same live-metadata-confirmed authoring discipline as backend/memory_evaluation/real_scenarios.py and the Phase D floor-plan sweep (experiments/reports/phase_d_floorplan_generalization_findings.md)",
+  "license": "mit",
+  "difficulty_tiers": {
+    "tier1_locate": "Single-step: locate a named object. Exercises object resolution only (rule_based.py's _goal_perceive).",
+    "tier2_pickup": "Two-step: navigate to and pick up a named object. Exercises navigate/pickup (rule_based.py's _acquire).",
+    "tier3_store": "Multi-step: pick up an object and place it into a named receptacle, opening/closing it if needed. Exercises rule_based.py's _deposit() -- the code path with this project's own documented bug history (Phase A's closed-receptacle fix, the Phase 8.7 audit, and Phase D's non-openable-target bug)."
+  },
+  "tasks": [
+    {"task_id": "milo-v1-fp1-t1a", "scene": "FloorPlan1", "room_type": "kitchen", "difficulty_tier": "tier1_locate", "instruction": "Find the mug.", "goal": "find", "object": "mug", "target": null},
+    {"task_id": "milo-v1-fp1-t1b", "scene": "FloorPlan1", "room_type": "kitchen", "difficulty_tier": "tier1_locate", "instruction": "Find the tomato.", "goal": "find", "object": "tomato", "target": null},
+    {"task_id": "milo-v1-fp1-t2a", "scene": "FloorPlan1", "room_type": "kitchen", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the apple.", "goal": "pick_up", "object": "apple", "target": null},
+    {"task_id": "milo-v1-fp1-t2b", "scene": "FloorPlan1", "room_type": "kitchen", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the spoon.", "goal": "pick_up", "object": "spoon", "target": null},
+    {"task_id": "milo-v1-fp1-t3a", "scene": "FloorPlan1", "room_type": "kitchen", "difficulty_tier": "tier3_store", "instruction": "Put the bread away in the fridge.", "goal": "store", "object": "bread", "target": "fridge", "notes": "fridge is a confirmed openable container in this scene."},
+
+    {"task_id": "milo-v1-fp5-t1a", "scene": "FloorPlan5", "room_type": "kitchen", "difficulty_tier": "tier1_locate", "instruction": "Find the bowl.", "goal": "find", "object": "bowl", "target": null},
+    {"task_id": "milo-v1-fp5-t1b", "scene": "FloorPlan5", "room_type": "kitchen", "difficulty_tier": "tier1_locate", "instruction": "Find the kettle.", "goal": "find", "object": "kettle", "target": null},
+    {"task_id": "milo-v1-fp5-t2a", "scene": "FloorPlan5", "room_type": "kitchen", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the potato.", "goal": "pick_up", "object": "potato", "target": null},
+    {"task_id": "milo-v1-fp5-t2b", "scene": "FloorPlan5", "room_type": "kitchen", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the pan.", "goal": "pick_up", "object": "pan", "target": null},
+    {"task_id": "milo-v1-fp5-t3a", "scene": "FloorPlan5", "room_type": "kitchen", "difficulty_tier": "tier3_store", "instruction": "Put the mug away in the cabinet.", "goal": "store", "object": "mug", "target": "cabinet", "notes": "cabinet is a confirmed openable container in this scene."},
+
+    {"task_id": "milo-v1-fp201-t1a", "scene": "FloorPlan201", "room_type": "living_room", "difficulty_tier": "tier1_locate", "instruction": "Find the laptop.", "goal": "find", "object": "laptop", "target": null},
+    {"task_id": "milo-v1-fp201-t1b", "scene": "FloorPlan201", "room_type": "living_room", "difficulty_tier": "tier1_locate", "instruction": "Find the vase.", "goal": "find", "object": "vase", "target": null},
+    {"task_id": "milo-v1-fp201-t2a", "scene": "FloorPlan201", "room_type": "living_room", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the newspaper.", "goal": "pick_up", "object": "newspaper", "target": null},
+    {"task_id": "milo-v1-fp201-t2b", "scene": "FloorPlan201", "room_type": "living_room", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the pillow.", "goal": "pick_up", "object": "pillow", "target": null},
+    {"task_id": "milo-v1-fp201-t3a", "scene": "FloorPlan201", "room_type": "living_room", "difficulty_tier": "tier3_store", "instruction": "Put the remote control away in the drawer.", "goal": "store", "object": "remotecontrol", "target": "drawer", "notes": "drawer is a confirmed openable container in this scene."},
+
+    {"task_id": "milo-v1-fp301-t1a", "scene": "FloorPlan301", "room_type": "bedroom", "difficulty_tier": "tier1_locate", "instruction": "Find the alarm clock.", "goal": "find", "object": "alarmclock", "target": null},
+    {"task_id": "milo-v1-fp301-t1b", "scene": "FloorPlan301", "room_type": "bedroom", "difficulty_tier": "tier1_locate", "instruction": "Find the cell phone.", "goal": "find", "object": "cellphone", "target": null},
+    {"task_id": "milo-v1-fp301-t2a", "scene": "FloorPlan301", "room_type": "bedroom", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the boots.", "goal": "pick_up", "object": "boots", "target": null},
+    {"task_id": "milo-v1-fp301-t2b", "scene": "FloorPlan301", "room_type": "bedroom", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the CD.", "goal": "pick_up", "object": "cd", "target": null},
+    {"task_id": "milo-v1-fp301-t3a", "scene": "FloorPlan301", "room_type": "bedroom", "difficulty_tier": "tier3_store", "instruction": "Put the book away in the drawer.", "goal": "store", "object": "book", "target": "drawer", "notes": "drawer is a confirmed openable container in this scene. Phase D found this exact task fails at execution due to a real AI2-THOR placement/geometry limit (no room found for the object's bounding box inside this drawer), not a planner defect -- kept in v1.0 anyway since it is a genuine, reproducible ground-truth data point, not authoring error. See experiments/reports/phase_d_floorplan_generalization_findings.md section 5."},
+
+    {"task_id": "milo-v1-fp401-t1a", "scene": "FloorPlan401", "room_type": "bathroom", "difficulty_tier": "tier1_locate", "instruction": "Find the towel.", "goal": "find", "object": "towel", "target": null},
+    {"task_id": "milo-v1-fp401-t1b", "scene": "FloorPlan401", "room_type": "bathroom", "difficulty_tier": "tier1_locate", "instruction": "Find the candle.", "goal": "find", "object": "candle", "target": null},
+    {"task_id": "milo-v1-fp401-t2a", "scene": "FloorPlan401", "room_type": "bathroom", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the soap bar.", "goal": "pick_up", "object": "soapbar", "target": null},
+    {"task_id": "milo-v1-fp401-t2b", "scene": "FloorPlan401", "room_type": "bathroom", "difficulty_tier": "tier2_pickup", "instruction": "Pick up the toilet paper.", "goal": "pick_up", "object": "toiletpaper", "target": null},
+    {"task_id": "milo-v1-fp401-t3a", "scene": "FloorPlan401", "room_type": "bathroom", "difficulty_tier": "tier3_store", "instruction": "Put the spray bottle away on the shelf.", "goal": "store", "object": "spraybottle", "target": "shelf", "notes": "KNOWN LIMITATION, kept deliberately: shelf is a real receptacle in this scene but is NOT openable. rule_based.py's _deposit() currently inserts an open step for every tier3_store target regardless of whether it is openable (a real bug this dataset's own generation sweep found -- see docs/roadmap.md's '_deposit() assumes every store destination is openable' row and experiments/reports/phase_d_floorplan_generalization_findings.md section 4). Any planner using today's rule_based.py deposit logic is expected to fail this task until that bug is fixed; this is intentional ground truth, not a mistake."}
+  ]
+}