Section 11.8: Choosing a Simulator

Step 1: Define The Task

Choose one task and write the dominant risk: contact fidelity, throughput, visual sensors, model-based control, manipulation benchmark coverage, or ROS 2 integration.

task = "tabletop pushing"
dominant_risk = "contact fidelity"
requirements = {"contact fidelity", "asset import", "batch rollout", "camera rendering"}
score = {"MuJoCo": 3, "Isaac Lab": 4, "Genesis": 3}
choice = max(score, key=score.get)
print({"task": task, "dominant_risk": dominant_risk, "recommended_stack": choice})

Step 2: Set Weights

Choose weights that match the task. The weights should sum to 1.0 so the final score is interpretable, and one criterion should capture whether the evidence was actually measured on this task.

Step 3: Score Candidates

Score at least three simulators from 1 to 5 using the same criteria. Keep the candidates in one table so the comparison is co-computed.

scores = {
    "MuJoCo": {
        "physics_fit": 5,
        "throughput": 3,
        "sensor_stack": 2,
        "integration": 3,
        "maturity": 5,
        "validation_evidence": 5,
    },
    "Isaac Lab": {
        "physics_fit": 4,
        "throughput": 5,
        "sensor_stack": 5,
        "integration": 4,
        "maturity": 4,
        "validation_evidence": 4,
    },
    "Modern Gazebo": {
        "physics_fit": 3,
        "throughput": 2,
        "sensor_stack": 4,
        "integration": 5,
        "maturity": 4,
        "validation_evidence": 4,
    },
}

if isinstance(scores, list):
    print({"rows": len(scores), "first": scores[0] if scores else None})
elif isinstance(scores, dict):
    print({"fields": sorted(scores), "audit_ready": all(value not in (None, "") for value in scores.values())})
else:
    print({"value": scores})

Step 4: Rank And Explain

Compute a ranking and print the recommendation with the exact reason.

ranked = sorted(
    ((name, sum(weights[k] * vals[k] for k in weights)) for name, vals in scores.items()),
    key=lambda item: item[1],
    reverse=True,
)
print(ranked[0])

('MuJoCo', 4.1)

Step 5: Add A Currency Check

Add one maintenance note for each candidate: current, frontier, legacy, or deprecated. Then add one validation note that states which evidence artifact supports the score.

currency = {
    "MuJoCo": "current",
    "Isaac Lab": "current",
    "Modern Gazebo": "current",
}
validation_artifacts = {
    "MuJoCo": "friction sweep replay",
    "Isaac Lab": "camera randomization replay",
    "Modern Gazebo": "ROS 2 topic and controller log",
}
def audit_release_support(
    currency: dict[str, str], validation_artifacts: dict[str, str]
) -> tuple[dict[str, str], dict[str, str]]:
    assert currency.keys() == validation_artifacts.keys()
    return currency, validation_artifacts

currency, validation_artifacts = audit_release_support(currency, validation_artifacts)
def audit_release_support(
    currency: dict[str, str], validation_artifacts: dict[str, str]
) -> tuple[dict[str, str], dict[str, str]]:
    assert currency.keys() == validation_artifacts.keys()
    return currency, validation_artifacts

currency, validation_artifacts = audit_release_support(currency, validation_artifacts)
print(currency)
print(validation_artifacts)

{'MuJoCo': 'current', 'Isaac Lab': 'current', 'Modern Gazebo': 'current'}
{'MuJoCo': 'friction sweep replay', 'Isaac Lab': 'camera randomization replay', 'Modern Gazebo': 'ROS 2 topic and controller log'}