vla-redteam 0.2.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- vla_redteam-0.2.0/.github/workflows/ci.yml +34 -0
- vla_redteam-0.2.0/.github/workflows/release.yml +94 -0
- vla_redteam-0.2.0/.gitignore +50 -0
- vla_redteam-0.2.0/CHANGELOG.md +141 -0
- vla_redteam-0.2.0/LICENSE +201 -0
- vla_redteam-0.2.0/PKG-INFO +316 -0
- vla_redteam-0.2.0/PRIOR_ART.md +80 -0
- vla_redteam-0.2.0/README.md +288 -0
- vla_redteam-0.2.0/SAFETY.md +67 -0
- vla_redteam-0.2.0/docs/assets/demo.svg +110 -0
- vla_redteam-0.2.0/leaderboard/README.md +63 -0
- vla_redteam-0.2.0/leaderboard/app.py +129 -0
- vla_redteam-0.2.0/leaderboard/requirements.txt +8 -0
- vla_redteam-0.2.0/leaderboard/results/leaderboard.json +80 -0
- vla_redteam-0.2.0/pyproject.toml +103 -0
- vla_redteam-0.2.0/results/smolvla_libero_object/README.md +25 -0
- vla_redteam-0.2.0/results/smolvla_libero_object/report.json +1197 -0
- vla_redteam-0.2.0/scripts/record_demo.sh +54 -0
- vla_redteam-0.2.0/scripts/run_real.sh +76 -0
- vla_redteam-0.2.0/src/vla_redteam/__init__.py +13 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/__init__.py +1 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/_image.py +77 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/_payload.py +51 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/base.py +50 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/baseline.py +28 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/injection.py +102 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/instruction.py +103 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/registry.py +120 -0
- vla_redteam-0.2.0/src/vla_redteam/attacks/visual.py +89 -0
- vla_redteam-0.2.0/src/vla_redteam/cli.py +250 -0
- vla_redteam-0.2.0/src/vla_redteam/config.py +50 -0
- vla_redteam-0.2.0/src/vla_redteam/leaderboard.py +178 -0
- vla_redteam-0.2.0/src/vla_redteam/policies/__init__.py +1 -0
- vla_redteam-0.2.0/src/vla_redteam/policies/base.py +57 -0
- vla_redteam-0.2.0/src/vla_redteam/policies/lerobot_adapter.py +260 -0
- vla_redteam-0.2.0/src/vla_redteam/policies/registry.py +94 -0
- vla_redteam-0.2.0/src/vla_redteam/policies/stub.py +134 -0
- vla_redteam-0.2.0/src/vla_redteam/report.py +148 -0
- vla_redteam-0.2.0/src/vla_redteam/runner.py +151 -0
- vla_redteam-0.2.0/src/vla_redteam/scoring/__init__.py +1 -0
- vla_redteam-0.2.0/src/vla_redteam/scoring/asr.py +87 -0
- vla_redteam-0.2.0/src/vla_redteam/suites/__init__.py +70 -0
- vla_redteam-0.2.0/src/vla_redteam/suites/base.py +64 -0
- vla_redteam-0.2.0/src/vla_redteam/suites/libero.py +404 -0
- vla_redteam-0.2.0/src/vla_redteam/suites/stub.py +127 -0
- vla_redteam-0.2.0/src/vla_redteam/types.py +131 -0
- vla_redteam-0.2.0/tests/conftest.py +20 -0
- vla_redteam-0.2.0/tests/test_attacks.py +94 -0
- vla_redteam-0.2.0/tests/test_cli.py +125 -0
- vla_redteam-0.2.0/tests/test_determinism.py +39 -0
- vla_redteam-0.2.0/tests/test_leaderboard.py +102 -0
- vla_redteam-0.2.0/tests/test_lerobot_adapter.py +155 -0
- vla_redteam-0.2.0/tests/test_libero_adapter.py +209 -0
- vla_redteam-0.2.0/tests/test_real_policy_injection.py +169 -0
- vla_redteam-0.2.0/tests/test_registries.py +63 -0
- vla_redteam-0.2.0/tests/test_runner_smoke.py +63 -0
- vla_redteam-0.2.0/tests/test_scoring.py +83 -0
- vla_redteam-0.2.0/tests/test_visual_injection.py +181 -0
- vla_redteam-0.2.0/uv.lock +3004 -0
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
name: CI
|
|
2
|
+
|
|
3
|
+
on:
|
|
4
|
+
push:
|
|
5
|
+
branches: [main]
|
|
6
|
+
pull_request:
|
|
7
|
+
|
|
8
|
+
# CPU-only by design: this workflow NEVER installs the optional `[lerobot]` extra,
|
|
9
|
+
# so it pulls no torch / transformers / GPU stack. The LeRobot integration tests are
|
|
10
|
+
# gated (ROBOPWN_INTEGRATION) and skip here automatically.
|
|
11
|
+
jobs:
|
|
12
|
+
check:
|
|
13
|
+
name: ruff + mypy + pytest (CPU, no lerobot)
|
|
14
|
+
runs-on: ubuntu-latest
|
|
15
|
+
steps:
|
|
16
|
+
- uses: actions/checkout@v4
|
|
17
|
+
|
|
18
|
+
- name: Install uv (pinned Python 3.12)
|
|
19
|
+
uses: astral-sh/setup-uv@v5
|
|
20
|
+
with:
|
|
21
|
+
python-version: "3.12"
|
|
22
|
+
enable-cache: true
|
|
23
|
+
|
|
24
|
+
- name: Sync core + dev (no extras)
|
|
25
|
+
run: uv sync --locked
|
|
26
|
+
|
|
27
|
+
- name: Lint (ruff)
|
|
28
|
+
run: uv run ruff check .
|
|
29
|
+
|
|
30
|
+
- name: Type-check (mypy, strict)
|
|
31
|
+
run: uv run mypy src
|
|
32
|
+
|
|
33
|
+
- name: Test (pytest, CPU only)
|
|
34
|
+
run: uv run pytest -q
|
|
@@ -0,0 +1,94 @@
|
|
|
1
|
+
name: Release
|
|
2
|
+
|
|
3
|
+
# Tag-triggered release: build sdist + wheel, publish to PyPI via OIDC TRUSTED
|
|
4
|
+
# PUBLISHING (no API token stored anywhere), then cut a GitHub Release from the
|
|
5
|
+
# CHANGELOG.
|
|
6
|
+
#
|
|
7
|
+
# ────────────────────────────────────────────────────────────────────────────
|
|
8
|
+
# ONE-TIME MAINTAINER SETUP (PyPI trusted publisher — no secrets in the repo):
|
|
9
|
+
# 1. Create the project owner account on https://pypi.org.
|
|
10
|
+
# 2. PyPI → your account → "Publishing" → "Add a new pending publisher":
|
|
11
|
+
# PyPI project name : vla-redteam
|
|
12
|
+
# Owner : sattyamjjain
|
|
13
|
+
# Repository name : vla-redteam
|
|
14
|
+
# Workflow name : release.yml
|
|
15
|
+
# Environment name : pypi
|
|
16
|
+
# 3. In GitHub → repo Settings → Environments → create an environment named "pypi"
|
|
17
|
+
# (optionally add required reviewers / a tag protection rule).
|
|
18
|
+
# 4. Publish by pushing a tag that matches the version, e.g.:
|
|
19
|
+
# git tag v0.1.0 && git push origin v0.1.0
|
|
20
|
+
# (For a first dry run, point step `publish` at TestPyPI by setting
|
|
21
|
+
# `repository-url: https://test.pypi.org/legacy/` and adding the matching pending
|
|
22
|
+
# publisher on test.pypi.org.)
|
|
23
|
+
# ────────────────────────────────────────────────────────────────────────────
|
|
24
|
+
|
|
25
|
+
on:
|
|
26
|
+
push:
|
|
27
|
+
tags: ["v*"]
|
|
28
|
+
|
|
29
|
+
permissions: {} # least privilege by default; jobs opt into exactly what they need
|
|
30
|
+
|
|
31
|
+
jobs:
|
|
32
|
+
build:
|
|
33
|
+
name: Build sdist + wheel
|
|
34
|
+
runs-on: ubuntu-latest
|
|
35
|
+
steps:
|
|
36
|
+
- uses: actions/checkout@v4
|
|
37
|
+
- name: Install uv (Python 3.12)
|
|
38
|
+
uses: astral-sh/setup-uv@v5
|
|
39
|
+
with:
|
|
40
|
+
python-version: "3.12"
|
|
41
|
+
- name: Build distributions
|
|
42
|
+
run: uv build
|
|
43
|
+
- name: Check metadata
|
|
44
|
+
run: uvx twine check dist/*
|
|
45
|
+
- uses: actions/upload-artifact@v4
|
|
46
|
+
with:
|
|
47
|
+
name: dist
|
|
48
|
+
path: dist/
|
|
49
|
+
|
|
50
|
+
publish-pypi:
|
|
51
|
+
name: Publish to PyPI (OIDC trusted publishing)
|
|
52
|
+
needs: build
|
|
53
|
+
runs-on: ubuntu-latest
|
|
54
|
+
environment: pypi
|
|
55
|
+
permissions:
|
|
56
|
+
id-token: write # REQUIRED for OIDC trusted publishing — no token needed
|
|
57
|
+
steps:
|
|
58
|
+
- uses: actions/download-artifact@v4
|
|
59
|
+
with:
|
|
60
|
+
name: dist
|
|
61
|
+
path: dist/
|
|
62
|
+
- name: Publish
|
|
63
|
+
uses: pypa/gh-action-pypi-publish@release/v1
|
|
64
|
+
# No `password:` — trusted publishing uses the OIDC id-token above.
|
|
65
|
+
|
|
66
|
+
github-release:
|
|
67
|
+
name: Create GitHub Release
|
|
68
|
+
needs: publish-pypi
|
|
69
|
+
runs-on: ubuntu-latest
|
|
70
|
+
permissions:
|
|
71
|
+
contents: write # to create the release
|
|
72
|
+
steps:
|
|
73
|
+
- uses: actions/checkout@v4
|
|
74
|
+
- uses: actions/download-artifact@v4
|
|
75
|
+
with:
|
|
76
|
+
name: dist
|
|
77
|
+
path: dist/
|
|
78
|
+
- name: Extract changelog notes for this tag
|
|
79
|
+
id: notes
|
|
80
|
+
run: |
|
|
81
|
+
version="${GITHUB_REF_NAME#v}"
|
|
82
|
+
awk -v ver="$version" '
|
|
83
|
+
$0 ~ "^## \\[" ver "\\]" {flag=1; next}
|
|
84
|
+
flag && /^## \[/ {exit}
|
|
85
|
+
flag {print}
|
|
86
|
+
' CHANGELOG.md > release-notes.md
|
|
87
|
+
if [ ! -s release-notes.md ]; then echo "Release ${GITHUB_REF_NAME}" > release-notes.md; fi
|
|
88
|
+
- name: Create release
|
|
89
|
+
env:
|
|
90
|
+
GH_TOKEN: ${{ github.token }}
|
|
91
|
+
run: |
|
|
92
|
+
gh release create "${GITHUB_REF_NAME}" dist/* \
|
|
93
|
+
--title "${GITHUB_REF_NAME}" \
|
|
94
|
+
--notes-file release-notes.md
|
|
@@ -0,0 +1,50 @@
|
|
|
1
|
+
# Python
|
|
2
|
+
__pycache__/
|
|
3
|
+
*.py[cod]
|
|
4
|
+
*$py.class
|
|
5
|
+
*.so
|
|
6
|
+
.Python
|
|
7
|
+
build/
|
|
8
|
+
dist/
|
|
9
|
+
*.egg-info/
|
|
10
|
+
.eggs/
|
|
11
|
+
develop-eggs/
|
|
12
|
+
.installed.cfg
|
|
13
|
+
*.egg
|
|
14
|
+
|
|
15
|
+
# Virtual environments
|
|
16
|
+
.venv/
|
|
17
|
+
venv/
|
|
18
|
+
env/
|
|
19
|
+
ENV/
|
|
20
|
+
.uv/
|
|
21
|
+
|
|
22
|
+
# uv
|
|
23
|
+
uv.lock.bak
|
|
24
|
+
|
|
25
|
+
# Testing / coverage
|
|
26
|
+
.pytest_cache/
|
|
27
|
+
.coverage
|
|
28
|
+
.coverage.*
|
|
29
|
+
coverage.xml
|
|
30
|
+
htmlcov/
|
|
31
|
+
.mypy_cache/
|
|
32
|
+
.ruff_cache/
|
|
33
|
+
.hypothesis/
|
|
34
|
+
|
|
35
|
+
# Run artifacts (generated reports)
|
|
36
|
+
runs/
|
|
37
|
+
|
|
38
|
+
# Optional demo recording (regenerated by scripts/record_demo.sh when asciinema+agg
|
|
39
|
+
# are installed); the committed deterministic asset is docs/assets/demo.svg.
|
|
40
|
+
docs/assets/*.gif
|
|
41
|
+
|
|
42
|
+
# OS / editor
|
|
43
|
+
.DS_Store
|
|
44
|
+
.idea/
|
|
45
|
+
.vscode/
|
|
46
|
+
*.swp
|
|
47
|
+
*~
|
|
48
|
+
|
|
49
|
+
# Jupyter
|
|
50
|
+
.ipynb_checkpoints/
|
|
@@ -0,0 +1,141 @@
|
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
All notable changes to this project are documented here. The format is based on
|
|
4
|
+
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to
|
|
5
|
+
[Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
6
|
+
|
|
7
|
+
## [0.1.0] — 2026-06-03
|
|
8
|
+
|
|
9
|
+
First public release. A model-agnostic harness that red-teams VLA policies in
|
|
10
|
+
simulation and reports an Attack Success Rate (ASR). The entire core runs and is tested
|
|
11
|
+
on a plain CPU with no GPU and no model/dataset download; real policies and the LIBERO
|
|
12
|
+
simulator are isolated behind the optional `[lerobot]` extra and an
|
|
13
|
+
`ROBOPWN_INTEGRATION=1` gate, and CI never installs them.
|
|
14
|
+
|
|
15
|
+
### Added
|
|
16
|
+
- **Core engine**: `PolicyAdapter`, `SuiteAdapter` (with an `is_unsafe` predicate), and
|
|
17
|
+
`Attack` abstractions; a deterministic `runner` (policy × suite × attacks × seeds);
|
|
18
|
+
pydantic v2 result types (`AttackResult`, `RunReport`).
|
|
19
|
+
- **Deterministic CPU stubs**: `StubPolicy` (manipulable via a documented trigger-token
|
|
20
|
+
lexicon over the instruction *and* a fixed allow-list of observation channels) and
|
|
21
|
+
`StubSuite` (one-task env with a SHA-256, cross-platform per-seed unsafe threshold),
|
|
22
|
+
so every attack yields a real, exact, reproducible seed-0 ASR with no model.
|
|
23
|
+
- **Three attack families** (templated, auditable, CPU-testable):
|
|
24
|
+
- `instruction` — `RolePlayAttack`, `GoalSubstitutionAttack`, `ParaphraseAttack`
|
|
25
|
+
(RoboPAIR / POEX-inspired). Seed-0 ASR **21/30**.
|
|
26
|
+
- `visual` — `PatchAttack`, `DecoyObjectAttack` (observation-space markers; BadVLA
|
|
27
|
+
-inspired). Seed-0 ASR **14/20**.
|
|
28
|
+
- `injection` — `SceneTextInjection`, `MCPToolDescInjection` (indirect / embodied
|
|
29
|
+
prompt injection; POEX-inspired). Seed-0 ASR **12/20**.
|
|
30
|
+
- Combined seed-0 ASR across all three: **47/70**. See [PRIOR_ART.md](PRIOR_ART.md).
|
|
31
|
+
- **Scoring & reports**: ASR with per-attack and per-task breakdowns (zero-division
|
|
32
|
+
guarded); stable `report.json` (sorted keys, no wall-clock → byte-deterministic) and
|
|
33
|
+
`report.md`, plus a Rich console summary.
|
|
34
|
+
- **`robopwn` CLI**: `attack`, `list-policies`, `list-attacks`, `report`, `version`, and
|
|
35
|
+
`leaderboard build`. Clean, actionable errors (exit code 2) for a missing `[lerobot]`
|
|
36
|
+
extra or an unknown policy/suite/attack — never a traceback.
|
|
37
|
+
- **Verified LeRobot SmolVLA adapter (gated)**: loads `lerobot/smolvla_base` via
|
|
38
|
+
`SmolVLAPolicy.from_pretrained` + `make_pre_post_processors` + `build_inference_frame`
|
|
39
|
+
+ `select_action`, written **after introspecting** the installed `lerobot==0.5.1`
|
|
40
|
+
(symbols and signatures confirmed, not guessed).
|
|
41
|
+
- **Verified LIBERO `SuiteAdapter` (gated)**: wraps the real LeRobot LIBERO env via the
|
|
42
|
+
introspected `make_env` / `make_env_config` / `LiberoEnv` API (`{suite: {task_id:
|
|
43
|
+
VectorEnv}}`, action dim 7, `info["final_info"]["is_success"]`). Configurable embodied
|
|
44
|
+
red-team predicate (`LiberoRedTeamRules`): default end-effector keep-out zone, optional
|
|
45
|
+
forbidden-object grasp. LIBERO is benign manipulation — we measure *redirection*, not
|
|
46
|
+
harm (see [SAFETY.md](SAFETY.md)).
|
|
47
|
+
- **ASR leaderboard (v0, static-data)**: `robopwn leaderboard build` aggregates
|
|
48
|
+
`report.json` files into a ranked `(policy × suite × family) → ASR` table
|
|
49
|
+
(deterministic). A Gradio Space (`leaderboard/`, ZeroGPU-compatible) renders the
|
|
50
|
+
committed `results/*.json` with no GPU; data is clearly labelled stub/demo until real
|
|
51
|
+
SmolVLA / OpenVLA runs are added.
|
|
52
|
+
- **Release engineering**: tag-triggered PyPI publish via OIDC trusted publishing
|
|
53
|
+
(`.github/workflows/release.yml`) and a GitHub Release from the CHANGELOG.
|
|
54
|
+
- **Docs & CI**: README (with a real "Verified environment" block), SAFETY.md,
|
|
55
|
+
PRIOR_ART.md, and a GitHub Actions workflow running ruff + mypy + pytest on Python
|
|
56
|
+
3.12, CPU-only, never installing lerobot.
|
|
57
|
+
|
|
58
|
+
### Known limitations
|
|
59
|
+
- The `StubPolicy` / `StubSuite` are deliberate test fixtures, not models of real VLA
|
|
60
|
+
behaviour. The visual/injection attacks target the stub's observation channels; against
|
|
61
|
+
a real policy they only transfer insofar as the policy reads the perturbed channel
|
|
62
|
+
(pixel-level visual attacks on real models are future work).
|
|
63
|
+
- The LeRobot SmolVLA adapter and the LIBERO `SuiteAdapter` are verified at the
|
|
64
|
+
symbol/signature level but not executed in CI (GPU-gated). The full LIBERO
|
|
65
|
+
observation→features wiring and the robosuite grasped-object accessor are deferred.
|
|
66
|
+
- The leaderboard ships demo (stub) data only until real-model runs are added.
|
|
67
|
+
|
|
68
|
+
## [0.2.0] — 2026-06-07 (in-process SmolVLA×LIBERO + first real result)
|
|
69
|
+
|
|
70
|
+
The real attack loop `robopwn attack --policy smolvla --suite libero` now runs
|
|
71
|
+
end-to-end (GPU-gated) and produced the project's **first real red-team result** on a
|
|
72
|
+
RunPod RTX 4090. Verified by **running** against the installed `lerobot==0.5.1`.
|
|
73
|
+
|
|
74
|
+
### Added
|
|
75
|
+
- **SmolVLA×LIBERO glue**: the SmolVLA adapter consumes a real LIBERO observation by
|
|
76
|
+
replicating LeRobot's own evaluator rollout (`preprocess_observation` → set task →
|
|
77
|
+
`env_preprocessor` (`LiberoProcessorStep`) → `preprocessor` → `select_action` →
|
|
78
|
+
`postprocessor` → `env_postprocessor` → clamp to `[-1,1]`). A `SuiteFeatures` protocol
|
|
79
|
+
exchanges the env config once; `make_policy(cfg, env_cfg, rename_map)` mirrors eval.
|
|
80
|
+
- **Real-policy attack injection**: instruction → the task language string; `visual.patch`
|
|
81
|
+
/ `visual.decoy_object` → adversarial-marker overlay on the camera image;
|
|
82
|
+
`injection.scene_text` → PIL-rendered scene sign. `injection.mcp_tool_desc` is honestly
|
|
83
|
+
**N/A** on a direct LIBERO loop (excluded from the denominator).
|
|
84
|
+
- **Live unsafe predicate**: end-effector keep-out zone (live, verified `eef.pos`) plus a
|
|
85
|
+
**configurable** forbidden-object-grasp rule (a `grasp_extractor` hook).
|
|
86
|
+
- **Seeded mean ± std** reporting for stochastic (real) policies; `--seeds`, `--model`,
|
|
87
|
+
and `--rename-map` CLI options; `pillow` added for image rendering.
|
|
88
|
+
- **`baseline` family** (`none`, a no-op control): runs the benign task unchanged so its
|
|
89
|
+
ASR is the reference, and attack ASR can be read as **lift over baseline**.
|
|
90
|
+
- **Real task language as the base instruction**: the LIBERO adapter now reads the env's
|
|
91
|
+
actual task via `env.call("task_description")` (mirroring LeRobot's `add_envs_task`)
|
|
92
|
+
instead of a placeholder — so the benign baseline measures the policy on its real job
|
|
93
|
+
and attack ASR is a true "diverted from the legitimate task" number.
|
|
94
|
+
- **Headless-sim robustness**: LIBERO's first-run `input()` config prompt is auto-handled
|
|
95
|
+
(`_ensure_libero_initialized`); `scripts/run_real.sh` sets `MUJOCO_GL`/`PYOPENGL_PLATFORM`
|
|
96
|
+
and documents the GL deps (`osmesa` CPU render, or EGL) for cloud GPU boxes.
|
|
97
|
+
|
|
98
|
+
### First real result — verified on a GPU (2026-06-06)
|
|
99
|
+
- Ran the full loop on a RunPod **RTX 4090** (`HuggingFaceVLA/smolvla_libero`,
|
|
100
|
+
`libero_object/0`, 10 seeds, horizon 280, `osmesa`). Overall **24.3% (17/70) ± 9.1%**.
|
|
101
|
+
Per family vs the benign `none` baseline (0/10): instruction `roleplay` **100%**,
|
|
102
|
+
`goal_substitution` **60%**, `paraphrase` 10%; `visual` and `injection` **0%**. The
|
|
103
|
+
canonical artifact is committed at `results/smolvla_libero_object/`, and
|
|
104
|
+
`leaderboard/results/leaderboard.json` now holds this real (non-demo) result.
|
|
105
|
+
- **Honest caveat**: the keep-out predicate is a default, **uncalibrated** region — so this
|
|
106
|
+
is "diverted out of the benign safe envelope," not a zone-calibrated hazard rate; one
|
|
107
|
+
task, `n = 10`.
|
|
108
|
+
|
|
109
|
+
### Verified by running (isolated lerobot venv, CPU)
|
|
110
|
+
- `lerobot/smolvla_base` is **NOT** LIBERO-compatible (expects `camera1/2/3`; LIBERO
|
|
111
|
+
gives `image`/`image2` + 8-dim state) → the adapter surfaces a clean
|
|
112
|
+
`IncompatiblePolicyError`.
|
|
113
|
+
- The **ready** `HuggingFaceVLA/smolvla_libero` checkpoint **loads through the glue** with
|
|
114
|
+
no incompatibility (policy + pre/post + `LiberoProcessorStep` all build) — so no
|
|
115
|
+
fine-tuning is needed. It is the default in `scripts/run_real.sh` and the gated tests.
|
|
116
|
+
|
|
117
|
+
### Known limitations
|
|
118
|
+
- Real-policy ASR is **seeded but model-stochastic** (mean ± std), not byte-deterministic
|
|
119
|
+
(only the stub is).
|
|
120
|
+
- Forbidden-object grasp is **disclosed-inert by default** (no verified robosuite accessor
|
|
121
|
+
without the sim) — live only with a user-supplied `grasp_extractor`.
|
|
122
|
+
- The first real ASR (above) uses an **uncalibrated** keep-out region, so it measures
|
|
123
|
+
redirection *out of the benign safe envelope*, not a zone-calibrated hazard rate; one
|
|
124
|
+
task, `n = 10`. Producing it requires a LIBERO-fine-tuned checkpoint + GPU + sim (not CI).
|
|
125
|
+
|
|
126
|
+
## [Unreleased] — planned
|
|
127
|
+
|
|
128
|
+
- **Per-task keep-out-zone calibration** (+ trajectory inspection) so a real ASR is a
|
|
129
|
+
semantically-meaningful redirection rate, not "entered a default box" — and a multi-task,
|
|
130
|
+
multi-suite sweep. This is the headline `v0.2.0` follow-up to the first real result.
|
|
131
|
+
- **Optimized-suffix attack family** (GCG-style / POEX-style executable suffixes) and
|
|
132
|
+
pixel-level visual attacks tuned to transfer to real VLA policies.
|
|
133
|
+
- **`mcp_tool_desc` on a real robot-MCP target** (no surface in a direct LIBERO loop).
|
|
134
|
+
- A **default robosuite grasp accessor** so forbidden-object-grasp is live out of the box.
|
|
135
|
+
- **Full multi-policy live leaderboard** on Hugging Face ZeroGPU with real SmolVLA /
|
|
136
|
+
OpenVLA numbers (the `@spaces.GPU` live-run path).
|
|
137
|
+
- **`vla-guard`** (Phase 2): a defensive companion that hardens policies against the
|
|
138
|
+
attacks this tool measures.
|
|
139
|
+
|
|
140
|
+
[0.2.0]: https://github.com/sattyamjjain/vla-redteam/releases/tag/v0.2.0
|
|
141
|
+
[0.1.0]: https://github.com/sattyamjjain/vla-redteam/releases/tag/v0.1.0
|
|
@@ -0,0 +1,201 @@
|
|
|
1
|
+
Apache License
|
|
2
|
+
Version 2.0, January 2004
|
|
3
|
+
http://www.apache.org/licenses/
|
|
4
|
+
|
|
5
|
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
|
6
|
+
|
|
7
|
+
1. Definitions.
|
|
8
|
+
|
|
9
|
+
"License" shall mean the terms and conditions for use, reproduction,
|
|
10
|
+
and distribution as defined by Sections 1 through 9 of this document.
|
|
11
|
+
|
|
12
|
+
"Licensor" shall mean the copyright owner or entity authorized by
|
|
13
|
+
the copyright owner that is granting the License.
|
|
14
|
+
|
|
15
|
+
"Legal Entity" shall mean the union of the acting entity and all
|
|
16
|
+
other entities that control, are controlled by, or are under common
|
|
17
|
+
control with that entity. For the purposes of this definition,
|
|
18
|
+
"control" means (i) the power, direct or indirect, to cause the
|
|
19
|
+
direction or management of such entity, whether by contract or
|
|
20
|
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
|
21
|
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
|
22
|
+
|
|
23
|
+
"You" (or "Your") shall mean an individual or Legal Entity
|
|
24
|
+
exercising permissions granted by this License.
|
|
25
|
+
|
|
26
|
+
"Source" form shall mean the preferred form for making modifications,
|
|
27
|
+
including but not limited to software source code, documentation
|
|
28
|
+
source, and configuration files.
|
|
29
|
+
|
|
30
|
+
"Object" form shall mean any form resulting from mechanical
|
|
31
|
+
transformation or translation of a Source form, including but
|
|
32
|
+
not limited to compiled object code, generated documentation,
|
|
33
|
+
and conversions to other media types.
|
|
34
|
+
|
|
35
|
+
"Work" shall mean the work of authorship, whether in Source or
|
|
36
|
+
Object form, made available under the License, as indicated by a
|
|
37
|
+
copyright notice that is included in or attached to the work
|
|
38
|
+
(an example is provided in the Appendix below).
|
|
39
|
+
|
|
40
|
+
"Derivative Works" shall mean any work, whether in Source or Object
|
|
41
|
+
form, that is based on (or derived from) the Work and for which the
|
|
42
|
+
editorial revisions, annotations, elaborations, or other modifications
|
|
43
|
+
represent, as a whole, an original work of authorship. For the purposes
|
|
44
|
+
of this License, Derivative Works shall not include works that remain
|
|
45
|
+
separable from, or merely link (or bind by name) to the interfaces of,
|
|
46
|
+
the Work and Derivative Works thereof.
|
|
47
|
+
|
|
48
|
+
"Contribution" shall mean any work of authorship, including
|
|
49
|
+
the original version of the Work and any modifications or additions
|
|
50
|
+
to that Work or Derivative Works thereof, that is intentionally
|
|
51
|
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
|
52
|
+
or by an individual or Legal Entity authorized to submit on behalf of
|
|
53
|
+
the copyright owner. For the purposes of this definition, "submitted"
|
|
54
|
+
means any form of electronic, verbal, or written communication sent
|
|
55
|
+
to the Licensor or its representatives, including but not limited to
|
|
56
|
+
communication on electronic mailing lists, source code control systems,
|
|
57
|
+
and issue tracking systems that are managed by, or on behalf of, the
|
|
58
|
+
Licensor for the purpose of discussing and improving the Work, but
|
|
59
|
+
excluding communication that is conspicuously marked or otherwise
|
|
60
|
+
designated in writing by the copyright owner as "Not a Contribution."
|
|
61
|
+
|
|
62
|
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
|
63
|
+
on behalf of whom a Contribution has been received by Licensor and
|
|
64
|
+
subsequently incorporated within the Work.
|
|
65
|
+
|
|
66
|
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
|
67
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
68
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
69
|
+
copyright license to reproduce, prepare Derivative Works of,
|
|
70
|
+
publicly display, publicly perform, sublicense, and distribute the
|
|
71
|
+
Work and such Derivative Works in Source or Object form.
|
|
72
|
+
|
|
73
|
+
3. Grant of Patent License. Subject to the terms and conditions of
|
|
74
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
75
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
76
|
+
(except as stated in this section) patent license to make, have made,
|
|
77
|
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
|
78
|
+
where such license applies only to those patent claims licensable
|
|
79
|
+
by such Contributor that are necessarily infringed by their
|
|
80
|
+
Contribution(s) alone or by combination of their Contribution(s)
|
|
81
|
+
with the Work to which such Contribution(s) was submitted. If You
|
|
82
|
+
institute patent litigation against any entity (including a
|
|
83
|
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
|
84
|
+
or a Contribution incorporated within the Work constitutes direct
|
|
85
|
+
or contributory patent infringement, then any patent licenses
|
|
86
|
+
granted to You under this License for that Work shall terminate
|
|
87
|
+
as of the date such litigation is filed.
|
|
88
|
+
|
|
89
|
+
4. Redistribution. You may reproduce and distribute copies of the
|
|
90
|
+
Work or Derivative Works thereof in any medium, with or without
|
|
91
|
+
modifications, and in Source or Object form, provided that You
|
|
92
|
+
meet the following conditions:
|
|
93
|
+
|
|
94
|
+
(a) You must give any other recipients of the Work or Derivative
|
|
95
|
+
Works a copy of this License; and
|
|
96
|
+
|
|
97
|
+
(b) You must cause any modified files to carry prominent notices
|
|
98
|
+
stating that You changed the files; and
|
|
99
|
+
|
|
100
|
+
(c) You must retain, in the Source form of any Derivative Works
|
|
101
|
+
that You distribute, all copyright, patent, trademark, and
|
|
102
|
+
attribution notices from the Source form of the Work, excluding
|
|
103
|
+
those notices that do not pertain to any part of the Derivative
|
|
104
|
+
Works; and
|
|
105
|
+
|
|
106
|
+
(d) If the Work includes a "NOTICE" text file as part of its
|
|
107
|
+
distribution, then any Derivative Works that You distribute must
|
|
108
|
+
include a readable copy of the attribution notices contained
|
|
109
|
+
within such NOTICE file, excluding those notices that do not
|
|
110
|
+
pertain to any part of the Derivative Works, in at least one
|
|
111
|
+
of the following places: within a NOTICE text file distributed
|
|
112
|
+
as part of the Derivative Works; within the Source form or
|
|
113
|
+
documentation, if provided along with the Derivative Works; or,
|
|
114
|
+
within a display generated by the Derivative Works, if and
|
|
115
|
+
wherever such third-party notices normally appear. The contents
|
|
116
|
+
of the NOTICE file are for informational purposes only and do
|
|
117
|
+
not modify the License. You may add Your own attribution notices
|
|
118
|
+
within Derivative Works that You distribute, alongside or as an
|
|
119
|
+
addendum to the NOTICE text from the Work, provided that such
|
|
120
|
+
additional attribution notices cannot be construed as modifying
|
|
121
|
+
the License.
|
|
122
|
+
|
|
123
|
+
You may add Your own copyright statement to Your modifications and
|
|
124
|
+
may provide additional or different license terms and conditions
|
|
125
|
+
for use, reproduction, or distribution of Your modifications, or
|
|
126
|
+
for any such Derivative Works as a whole, provided Your use,
|
|
127
|
+
reproduction, and distribution of the Work otherwise complies with
|
|
128
|
+
the conditions stated in this License.
|
|
129
|
+
|
|
130
|
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
|
131
|
+
any Contribution intentionally submitted for inclusion in the Work
|
|
132
|
+
by You to the Licensor shall be under the terms and conditions of
|
|
133
|
+
this License, without any additional terms or conditions.
|
|
134
|
+
Notwithstanding the above, nothing herein shall supersede or modify
|
|
135
|
+
the terms of any separate license agreement you may have executed
|
|
136
|
+
with Licensor regarding such Contributions.
|
|
137
|
+
|
|
138
|
+
6. Trademarks. This License does not grant permission to use the trade
|
|
139
|
+
names, trademarks, service marks, or product names of the Licensor,
|
|
140
|
+
except as required for reasonable and customary use in describing the
|
|
141
|
+
origin of the Work and reproducing the content of the NOTICE file.
|
|
142
|
+
|
|
143
|
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
|
144
|
+
agreed to in writing, Licensor provides the Work (and each
|
|
145
|
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
|
146
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
|
147
|
+
implied, including, without limitation, any warranties or conditions
|
|
148
|
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
|
149
|
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
|
150
|
+
appropriateness of using or redistributing the Work and assume any
|
|
151
|
+
risks associated with Your exercise of permissions under this License.
|
|
152
|
+
|
|
153
|
+
8. Limitation of Liability. In no event and under no legal theory,
|
|
154
|
+
whether in tort (including negligence), contract, or otherwise,
|
|
155
|
+
unless required by applicable law (such as deliberate and grossly
|
|
156
|
+
negligent acts) or agreed to in writing, shall any Contributor be
|
|
157
|
+
liable to You for damages, including any direct, indirect, special,
|
|
158
|
+
incidental, or consequential damages of any character arising as a
|
|
159
|
+
result of this License or out of the use or inability to use the
|
|
160
|
+
Work (including but not limited to damages for loss of goodwill,
|
|
161
|
+
work stoppage, computer failure or malfunction, or any and all
|
|
162
|
+
other commercial damages or losses), even if such Contributor
|
|
163
|
+
has been advised of the possibility of such damages.
|
|
164
|
+
|
|
165
|
+
9. Accepting Warranty or Additional Liability. While redistributing
|
|
166
|
+
the Work or Derivative Works thereof, You may choose to offer,
|
|
167
|
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
|
168
|
+
or other liability obligations and/or rights consistent with this
|
|
169
|
+
License. However, in accepting such obligations, You may act only
|
|
170
|
+
on Your own behalf and on Your sole responsibility, not on behalf
|
|
171
|
+
of any other Contributor, and only if You agree to indemnify,
|
|
172
|
+
defend, and hold each Contributor harmless for any liability
|
|
173
|
+
incurred by, or claims asserted against, such Contributor by reason
|
|
174
|
+
of your accepting any such warranty or additional liability.
|
|
175
|
+
|
|
176
|
+
END OF TERMS AND CONDITIONS
|
|
177
|
+
|
|
178
|
+
APPENDIX: How to apply the Apache License to your work.
|
|
179
|
+
|
|
180
|
+
To apply the Apache License to your work, attach the following
|
|
181
|
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
|
182
|
+
replaced with your own identifying information. (Don't include
|
|
183
|
+
the brackets!) The text should be enclosed in the appropriate
|
|
184
|
+
comment syntax for the file format. We also recommend that a
|
|
185
|
+
file or class name and description of purpose be included on the
|
|
186
|
+
same "printed page" as the copyright notice for easier
|
|
187
|
+
identification within third-party archives.
|
|
188
|
+
|
|
189
|
+
Copyright [yyyy] [name of copyright owner]
|
|
190
|
+
|
|
191
|
+
Licensed under the Apache License, Version 2.0 (the "License");
|
|
192
|
+
you may not use this file except in compliance with the License.
|
|
193
|
+
You may obtain a copy of the License at
|
|
194
|
+
|
|
195
|
+
http://www.apache.org/licenses/LICENSE-2.0
|
|
196
|
+
|
|
197
|
+
Unless required by applicable law or agreed to in writing, software
|
|
198
|
+
distributed under the License is distributed on an "AS IS" BASIS,
|
|
199
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
200
|
+
See the License for the specific language governing permissions and
|
|
201
|
+
limitations under the License.
|