vla-evaluation-harness
✓ reproduced | ◇ integrated, awaiting first reproduction | · planned
One framework to evaluate any VLA model on any robot simulation benchmark.
Latest News
- [2026/09] Push-T example: a self-contained project that trains on its own environment and evaluates it with
vla_eval.evaluate during training. pip install vla-eval is all it needs, no clone of this repo.
- [2026/09] v0.6.0 released. Python API for training-time evaluation, Charliecloud runtime, and
docker.build.
- [2026/08] v0.5.0 released. RoboDojo, the RoboCasa/RoboCasa365 split, and the CPU render backend.
- [2026/07] v0.4.0 released. Recording on by default, pinned reproducible Docker rebuilds, DuoBench, and the LeRobot bridge below.
- [2026/07] 🤗 LeRobot bridge: serve any single-obs-step LeRobot
PreTrainedPolicy (π₀ / π₀.₅, GR00T N1.7, X-VLA, MolmoAct2, and the FastWAM / VLA-JEPA / LingBot-VA world models) with one config, pinned at v0.6.0. Four checkpoints reproduce their published LIBERO scores: π₀.₅ 100%, GR00T N1.7 99%, MolmoAct2 97%, VLA-JEPA 96%.
- [2026/06] v0.3.0 released. SQLite recording +
vla-eval merge, wandb/trackio tracking, and a watchdog for wedged benchmarks.
- [2026/05] v0.2.0 released. 18 benchmarks x 13 model servers, the largest open VLA evaluation matrix. Browse to get started.
Why vla-evaluation-harness?
| |
|---|
| Batch Parallel Evaluation | Episode sharding + batched GPU inference → 47× throughput (2 000 LIBERO episodes in 18 min on 1× H100). Details |
| Zero Setup | Benchmarks in Docker and model servers as single-file uv scripts, avoiding dependency conflicts. |
| AI-Assisted Integration | Shared agent skills for adding benchmarks, model servers, and running evaluations. |
| Leaderboard | The largest unified VLA comparison: 2,456 models × 18 benchmarks, aggregated from 2,087 papers. |
Motivation
VLA models are evaluated on LIBERO, CALVIN, SimplerEnv, ManiSkill, and others, but each benchmark has its own dependencies, observation format, and evaluation protocol. In practice, every research team ends up maintaining private eval forks per benchmark. Results diverge. Bug fixes don't propagate. No one tests under real-time conditions where the environment keeps moving during inference.
vla-evaluation-harness integrates the model once, integrates the benchmark once, and the full cross-evaluation matrix fills itself.
How: our abstraction layer fully decouples models from benchmarks.
- Benchmarks run inside Docker, giving exact reproducibility without dependency conflicts.
- Model servers are standalone uv scripts with inline dependency declarations and no manual setup.
See Architecture for how the pieces connect.
Installation
Requires Python 3.8+ and SQLite 3.24+ with JSON support for recording. A uv-managed
Python build includes a suitable SQLite; system Python builds may not.
pip install vla-eval
Or from source (pinned to the latest stable release):
git clone --branch v0.7.0 https://github.com/allenai/vla-evaluation-harness.git
cd vla-evaluation-harness
uv sync --python 3.11 --all-extras --dev --group leaderboard
Quick Start
Two terminals: one for the model server (GPU), one for the benchmark client.
# Terminal 1: model server (runs on host with GPU)
vla-eval serve --config configs/model_servers/db_cogact/libero.yaml
# Terminal 2: run evaluation (benchmark runs in Docker by default).
# Wait for the model server to finish loading first; ``GET /health`` returning HTTP 200 is the ready signal.
vla-eval run --config configs/benchmarks/libero/smoke_test.yaml
Results are saved to results/ as JSON. The benchmark runs inside Docker by default; pass --no-docker for local development.
No Docker daemon? --runtime charliecloud runs the same image without root; see Container runtimes.
For full evaluation (10 tasks x 50 episodes):
vla-eval run --config configs/benchmarks/libero/spatial.yaml
Other benchmarks and models follow the same pattern. Pick a benchmark and a compatible model server from configs/:
# SimplerEnv + X-VLA
vla-eval serve --config configs/model_servers/xvla/simpler_widowx.yaml
vla-eval run --config configs/benchmarks/simpler/widowx_vm.yaml
# CALVIN + DB-CogACT
vla-eval serve --config configs/model_servers/db_cogact/calvin.yaml
vla-eval run --config configs/benchmarks/calvin/eval.yaml
# LIBERO + π₀.₅ via 🤗 LeRobot (works for any LeRobot PreTrainedPolicy checkpoint)
vla-eval serve --config configs/model_servers/lerobot/pi05_libero.yaml
vla-eval run --config configs/benchmarks/libero/object.yaml
Each benchmark and model server directory has a README with setup details, supported configs, and Docker image info. See Reproduction Reports for verified scores.
Need faster runs? See Batch Parallel Evaluation for up to 47x throughput.
From Python (evaluate while training)
The same run is one function call, with the model served from the calling process:
import vla_eval
results = vla_eval.evaluate(MyModelServer(model), "configs/benchmarks/libero/smoke_test.yaml")
print(results[0]["mean_success"])
Python API documents the arguments. examples/pusht_train_eval
is a complete training script that does this every N steps on an environment vla-eval does not ship.
Batch Parallel Evaluation
A full evaluation takes hours sequentially. Two layers of parallelism bring this down to minutes:
Episode sharding splits (task, episode) pairs across N independent processes (RFC-0006). Each shard connects to the same model server, where a BatchPredictModelServer batches their inference requests into a single forward pass. The two axes multiply together.
Episode Sharding (environment parallelism)
# Option A: use the helper script (launches all shards + auto-exports)
./scripts/run_sharded.sh -c configs/benchmarks/libero/spatial.yaml -n 50
# Option B: manual launch
EVAL_ID=$(uuidgen)
vla-eval run -c configs/benchmarks/libero/spatial.yaml --eval-id "$EVAL_ID" --shard-id 0 --num-shards 4 &
vla-eval run -c configs/benchmarks/libero/spatial.yaml --eval-id "$EVAL_ID" --shard-id 1 --num-shards 4 &
# ... (each shard is a separate process)
wait
vla-eval export "results/recording-$EVAL_ID.sqlite"
Shards claim work items from a queue in the shared recording DB (--no-save: fixed round-robin split). A shard whose first three episodes all error exits with code 3; --requeue-unhealthy gives its items to the other shards. Re-run a shard with the same eval ID to resume its unfinished items; export keeps the last committed attempt per item and lists unfinished ones.
Benchmark containers run as root by default, so output_dir ends up root-owned and the host-side vla-eval export fails with Permission denied. Set docker.user: host in the eval YAML (or an explicit "<uid>:<gid>") to run the containers as the invoking user.
Batch Model Server (GPU parallelism)
Enable batching in the model server config by setting max_batch_size > 1:
args:
max_batch_size: 16 # max observations per GPU forward pass (>1 enables batching)
max_wait_time: 0.05 # seconds to wait before dispatching a partial batch
Tuning & Combined Effect
We tune parallelism via a demand/supply methodology: demand λ(N) measures environment throughput as a function of shards, supply μ(B) measures model throughput as a function of batch size. The operating point satisfies λ(N) < 80% · μ(B*) to prevent queue buildup.
Sharding and batching multiply together (DB-CogACT 7B, LIBERO Spatial, 1× H100-80GB):
| Sequential | Batch Parallel (50 shards, B=16) |
|---|
| Wall-clock | ~14 h | ~18 min |
| Throughput | ~11 obs/s | ~486 obs/s |
2 000 episodes, 47× faster. The included benchmarking tools (experiments/bench_demand.py, experiments/bench_supply.py) measure λ and μ for any model + benchmark combination. See the Tuning Guide for worked examples and max_wait_time derivation.
Docker Images
All benchmark environments are packaged as standalone Docker images, based on base unless noted.
🔒 = build-locally only; the Dockerfile gates the build behind a licence opt-in (docker/build.sh <name> --accept-license <name>) and the image isn't published to ghcr.io.
Pull (recommended):
docker pull ghcr.io/allenai/vla-evaluation-harness/libero:latest
Build locally (see docker/build.sh):
docker/build.sh # build all (gated images skipped)
docker/build.sh libero # build one
docker/build.sh behavior1k --accept-license behavior1k # build a gated image
Build from your own Dockerfile: set docker.build: <context> (compose semantics, optional dockerfile:) in the benchmark config. The image is built when missing, or on --build. The Push-T example uses it.
Observability
Two systems capture eval data. Both key on eval_id so recordings and tracker runs stay linked.
Recording
Benchmark entries persist episode results and step rows by default, with videos off. Add a top-level recording: key only when you want to override that policy; use --record-video or set record_video: true when you want per-episode mp4s.
benchmarks:
- benchmark: ...
recording:
record_video: true
video_fps: 10
Recording writes <output_dir>/recording-<eval_id>.sqlite: per-step rows, episode results, and eval metadata. vla-eval export materializes per-episode JSONL + aggregate JSON from the DB. Single-shard runs auto-export; sharded runs call vla-eval export once after all shards exit. --no-save skips recording entirely.
Use vla-eval export DB [-o DIR] to export recorded results (default: DB parent). All writers need write access to the DB and its directory; shared storage must support SQLite locking and sync.
When filename_stem is omitted, per-episode artifacts use a benchmark-scoped path:
{benchmark_safe_name}/task{task_idx:04d}_ep{episode_id:04d}_{status}. Custom stems can still reference serializable task fields such as {name} plus {status}.
Tracking (wandb / trackio)
Mirror aggregate metrics to a remote dashboard:
tracking:
report_to: wandb # "wandb" | ["wandb", "trackio"] | "all" | "none"
The harness injects id=<eval_id> + resume="allow" so live (vla-eval run) and export (vla-eval export) paths converge on the same run. All other settings, including project, entity, and API key, come from the backend's native env vars (WANDB_*, TRACKIO_*). See the W&B env reference for details. Install the backend yourself: pip install wandb / pip install trackio.
Under sharding, aggregate emission defers to vla-eval export; per-episode tracking is live-path only.
Documentation
| Document | Description |
|---|
| Architecture | Component descriptions, protocol, episode flow, configuration |
| Render Backends | Running the simulator on the CPU (--render cpu) to free the GPU for the model |
| Container runtimes | Docker vs Charliecloud (--runtime charliecloud, no daemon, no root) |
| Python API | evaluate() / run() / serve_background() for calling the harness from a training script |
| Push-T example | Train with LeRobot and evaluate with vla-eval during training, on your own environment |
| Tuning Guide | Measuring λ / μ and deriving max_wait_time for batch-parallel runs |
| Contributing | Dev setup, adding benchmarks/models, PR workflow |
| Reproduction Reports | Per-model evaluation results and reproducibility verdicts |
| RFCs | Design proposals with rationale and status tracking |
|
Contributing
See CONTRIBUTING.md for dev setup and PR workflow.
PRs for any 🔜 item in the support matrix are welcome.
Citation
If you find this work useful, please cite:
@article{choi2026vlaeval,
title={vla-eval: A Unified Evaluation Harness for Vision-Language-Action Models},
author={Choi, Suhwan and Lee, Yunsung and Park, Yubeen and Kim, Chris Dongjoo and Krishna, Ranjay and Fox, Dieter and Yu, Youngjae},
journal={arXiv preprint arXiv:2603.13966},
year={2026}
}
License
Apache 2.0
Agent skills
The skills in skills/ support Codex, Claude Code, and other
npx skills clients. Use them from a harness
checkout; installing a skill does not install the harness or simulator dependencies.
# Discover available skills from this checkout.
npx skills add . --list
# Install into another project from a local checkout.
npx skills add /path/to/vla-evaluation-harness --agent codex claude-code --skill '*'
# Install a selected skill from the public repository after this change lands.
npx skills add worv-ai/vla-evaluation-harness-public --agent codex --skill add-benchmark
.agents/skills and .claude/skills point to the same source. Shared repository
instructions live in AGENTS.md; CLAUDE.md imports them for Claude Code.
