Expose RL benchmark warm-up in Python API

Keep typed training and play requests equivalent to the CLI added by PR isaac-sim#6474. Translate warm-up steps through dispatch so callers do not need backend-specific arguments.

Author: AntoineRichard

source/isaaclab/isaaclab/benchmark/api.py

@@ -153,6 +153,7 @@ class BenchmarkTrainingRequest:
         num_envs: Number of parallel environments.
         seed: Environment and agent seed.
         max_iterations: Maximum training iterations.
+        warmup_steps: Number of initial environment steps excluded from environment-step timing.
         ray_proc_id: Ray worker process identifier.
         video: Whether to record training videos.
         video_length: Recorded video length [steps].
@@ -182,6 +183,7 @@ class BenchmarkTrainingRequest:
     num_envs: int | None = None
     seed: int | None = None
     max_iterations: int | None = None
+    warmup_steps: int = 0
     ray_proc_id: int | None = None
     video: bool = False
     video_length: int = 200
@@ -221,6 +223,7 @@ class BenchmarkPlayRequest:
         agent: Optional task agent configuration entry point.
         num_envs: Number of parallel environments.
         num_frames: Number of measured inference steps.
+        warmup_steps: Number of initial environment steps excluded from environment-step timing.
         seed: Environment seed.
         measure_synchronized_step_breakdown: Whether to collect serialized synchronized
             environment/simulation step diagnostics.
@@ -237,6 +240,7 @@ class BenchmarkPlayRequest:
     agent: str | None = None
     num_envs: int | None = None
     num_frames: int = 100
+    warmup_steps: int = 0
     seed: int | None = None
     measure_synchronized_step_breakdown: bool = False
     presets: tuple[str, ...] = field(default_factory=tuple)

source/isaaclab/isaaclab/benchmark/dispatch.py

@@ -135,6 +135,7 @@ def _request_argv(request: BenchmarkRequest) -> list[str]:
     elif request.workflow == "training":
         _append_value(argv, "--agent", request.agent)
         _append_value(argv, "--max_iterations", request.max_iterations)
+        _append_value(argv, "--warmup_steps", request.warmup_steps)
         _append_value(argv, "--ray-proc-id", request.ray_proc_id)
         if request.video:
             argv.append("--video")
@@ -160,6 +161,7 @@ def _request_argv(request: BenchmarkRequest) -> list[str]:
         _append_value(argv, "--checkpoint", request.checkpoint)
         _append_value(argv, "--agent", request.agent)
         _append_value(argv, "--num_frames", request.num_frames)
+        _append_value(argv, "--warmup_steps", request.warmup_steps)
 
     if getattr(request, "measure_synchronized_step_breakdown", False):
         argv.append("--measure_synchronized_step_breakdown")

source/isaaclab/test/benchmark/test_api.py

@@ -17,6 +17,7 @@
 from isaaclab.benchmark import (
     BenchmarkLauncherConfig,
     BenchmarkOutputConfig,
+    BenchmarkPlayRequest,
     BenchmarkResult,
     BenchmarkRuntimeRequest,
     BenchmarkTrainingRequest,
@@ -35,6 +36,7 @@ def test_training_request_builds_complete_cli() -> None:
         num_envs=64,
         seed=7,
         max_iterations=10,
+        warmup_steps=12,
         ray_proc_id=4,
         video=True,
         video_length=100,
@@ -70,6 +72,8 @@ def test_training_request_builds_complete_cli() -> None:
         "custom_agent",
         "--max_iterations",
         "10",
+        "--warmup_steps",
+        "12",
         "--ray-proc-id",
         "4",
         "--video",
@@ -123,6 +127,14 @@ def test_runtime_request_uses_runtime_defaults() -> None:
     ]
 
 
+def test_play_request_includes_warmup_steps() -> None:
+    request = BenchmarkPlayRequest(backend="rsl_rl", task="Isaac-Cartpole-Direct", warmup_steps=12)
+
+    argv = dispatch._request_argv(request)
+
+    assert argv[argv.index("--warmup_steps") + 1] == "12"
+
+
 @pytest.mark.parametrize("workflow", ["runtime", "startup"])
 def test_task_workflow_help_does_not_require_task(workflow: str, monkeypatch, capsys) -> None:
     entrypoint = importlib.import_module(f"isaaclab.benchmark.entrypoints.{workflow}")