# RLBench (/docs/evaluation/benchmark-hub/rlbench)



## About [#about]

RLBench external benchmark task manifest for CoppeliaSim/PyRep robot manipulation rollouts.

## Metrics [#metrics]

RLBench external benchmark task manifest for CoppeliaSim/PyRep robot manipulation rollouts.

### Task YAML [#task-yaml]

`worldfoundry/data/benchmarks/tasks/external/rlbench.yaml`

### Primary [#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.                                   |
| `reward`            | Sparse or shaped reward accumulated over RLBench tabletop manipulation episode horizons. |
| `normalized_return` | RLBench return normalized to \[0, 1] against demonstration or success benchmarks.        |

## Environment setup [#environment-setup]

### Runtime environment [#runtime-environment]

* **Conda env:** `needs-rlbench-pyrep-coppeliasim-env`.
* **Separate setup:** this benchmark expects a dedicated conda env rather than the unified runtime.
* RLBench 需要 CoppeliaSim/PyRep 组合环境；当前只记录 env vars 和 task manifest，不启动 simulator。

### Evaluation data [#evaluation-data]

* RLBench is simulator-backed; WorldFoundry normalizes policy rollout results while CoppeliaSim/PyRep execution remains explicit.
* Candidate model outputs: set `WORLDFOUNDRY_RLBENCH_RESULTS_PATH` to the generated-video or rollout artifact root.

### Checkpoints & assets [#checkpoints--assets]

* This benchmark is primarily rollout- or official-result-driven; see catalog `requires` and the task YAML for policy/simulator checkpoints.

### Key environment variables [#key-environment-variables]

* `WORLDFOUNDRY_RLBENCH_ROOT`
* `COPPELIASIM_ROOT`
* `WORLDFOUNDRY_RLBENCH_RESULTS_PATH`

### Verify setup [#verify-setup]

```bash
PYTHONPATH=. ${WORLDFOUNDRY_CONDA_ROOT}/worldplay/bin/python -m worldfoundry.evaluation.tasks.execution.orchestration.runtime_preflight --profile rlbench --manifest worldfoundry/data/benchmarks/runtime_profiles/official/rlbench.yaml --output-dir tmp/worldfoundry_preflight/rlbench --json
```

## Run Evaluation [#run-evaluation]

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

### Set Candidate Artifacts [#set-candidate-artifacts]

```bash
export WORLDFOUNDRY_GENERATED_ARTIFACT_DIR=/path/to/generated/artifacts
```

### Import Official Results [#import-official-results]

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

### Score Generated Artifacts [#score-generated-artifacts]

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

### Direct In-Tree Runner [#direct-in-tree-runner]

```bash
worldfoundry-eval zoo benchmark-run --benchmark-id rlbench --mode official-validation --official-results-path <official_results.json> --output-dir <out> --json
```

## Requirements [#requirements]

### Inputs [#inputs]

* Official RLBench repository checkout
* CoppeliaSim and PyRep runtime
* RLBench task assets
* Benchmark-specific policy checkpoint

### Environment [#environment]

* `WORLDFOUNDRY_RLBENCH_RESULTS_PATH`

### Outputs [#outputs]

* `scorecard.json`
* `raw_results.jsonl`

[← Benchmark Hub](/docs/evaluation/benchmark-hub)