RoboTwin 2.0

Normalizer

RoboTwin 2.0 — metrics, requirements, and run commands.

Last updated: Jul 11, 2026

On this page

About Metrics Task YAML Primary Environment setup Runtime environment Evaluation data Checkpoints & assets Key environment variables Run Evaluation Set Candidate Artifacts Import Official Results Score Generated Artifacts Direct In-Tree Runner Requirements Inputs Environment Outputs

About

RoboTwin 2.0 is a bimanual manipulation benchmark with diverse tasks and sim-to-real oriented evaluation.

Metrics

RoboTwin 2.0 external benchmark task manifest for bimanual embodied manipulation policy rollouts.

Task YAML

worldfoundry/data/benchmarks/tasks/external/robotwin.yaml

Primary

success_rate

Metric	Focus
`success_rate`	Share of episodes or tasks that complete successfully.
`task_success`	Share of episodes or tasks that complete successfully.
`episode_success`	Share of episodes or tasks that complete successfully.
`bimanual_task_success`	Share of episodes or tasks that complete successfully.
`domain_randomization_success`	Share of episodes or tasks that complete successfully.
`clean_success`	Share of episodes or tasks that complete successfully.
`randomized_success`	Share of episodes or tasks that complete successfully.
`reward`	Cumulative reward on RoboTwin dual-arm manipulation episodes in the official benchmark suite.

Environment setup

Runtime environment

Conda env: docker:robotwin.
Separate setup: defaults to the unified WorldFoundry env (needs_new_env: false); a benchmark-only conda env is usually not required.
Embodied simulator dependencies are isolated in the VLA evaluation harness Docker image; WorldFoundry does not ship a RoboTwin conda setup script.
Host-side WorldFoundry only needs Docker access, the policy/model server, and mounted assets/results.

Evaluation data

Candidate model outputs: set WORLDFOUNDRY_ROBOTWIN_RESULTS_PATH to the generated-video or rollout artifact root.
${WORLDFOUNDRY_HFD_DATASET_ROOT}/TianxingChen__RoboTwin2.0
${WORLDFOUNDRY_HFD_DATASET_ROOT}/lerobot__robotwin_unified

Checkpoints & assets

Base-model / metric dependencies: robotwin_dataset_assets.
Default metric-checkpoint paths and override env vars are listed in the local assets guide.

Key environment variables

WORLDFOUNDRY_ROBOTWIN_RESULTS_PATH

Run Evaluation

After preparing the assets and candidate artifacts listed on this page, use the public run entry points below.

Set Candidate Artifacts

export WORLDFOUNDRY_GENERATED_ARTIFACT_DIR=/path/to/generated/artifacts

Import Official Results

worldfoundry-eval zoo benchmark-run \
  --benchmark-id robotwin \
  --mode normalizer \
  --official-results-path /path/to/official/results-or-report \
  --generated-artifact-dir "${WORLDFOUNDRY_GENERATED_ARTIFACT_DIR}" \
  --output-dir tmp/robotwin/normalizer \
  --json

Score Generated Artifacts

worldfoundry-eval zoo benchmark-run \
  --benchmark-id robotwin \
  --mode official-run \
  --generated-artifact-dir "${WORLDFOUNDRY_GENERATED_ARTIFACT_DIR}" \
  --output-dir tmp/robotwin/official-run \
  --json

Direct In-Tree Runner

worldfoundry-eval embodied run --config worldfoundry/data/benchmarks/eval_configs/embodied/robotwin/eval.yaml

Requirements

Inputs

Official benchmark runtime or preflight validation

Environment

WORLDFOUNDRY_ROBOTWIN_RESULTS_PATH

Outputs

scorecard.json
benchmark_contract.json
task_manifest.json
official_command_manifest.json
simulator_validation.json
raw_metric_table.jsonl
evaluation/per_sample_metrics.jsonl

← Benchmark Hub

RoboTwin 2.0

Related