Aggregate open benchmarking work

docs/source/api/lab/isaaclab.benchmark.rst

@@ -34,9 +34,41 @@ isaaclab.benchmark
       MethodBenchmarkDefinition
       MethodBenchmarkRunner
       MethodBenchmarkRunnerConfig
+
+   .. rubric:: Measurements and metadata
+
+   .. autosummary::
+
+      BooleanMeasurement
+      DictMeasurement
+      DictMetadata
+      FloatMetadata
+      IntMetadata
+      ListMeasurement
+      Measurement
+      MetadataBase
+      SingleMeasurement
+      StatisticalMeasurement
+      StringMetadata
+      TestPhase
+
+   .. rubric:: Result schema
+
+   .. autosummary::
+
+      EnvironmentStepTiming
+      Learning
+      LearningCurve
+      MeanStd
       PlayBundle
+      Resources
+      RunConfig
+      RunIdentity
+      Runtime
       RuntimeBundle
       StartupBundle
+      StartupPhase
+      StartupTime
       TrainingBundle
 
 

docs/source/overview/reinforcement-learning/performance_benchmarks.rst

@@ -125,27 +125,10 @@ CPU: Intel(R) Xeon(R) Platinum 8362 CPU @ 2.80GHz
 +-------------------------------------+-------------------+--------------+-------------------+--------------------+
 
 
-Benchmark Scripts
------------------
-
-For ease of reproducibility, we provide benchmarking scripts available at ``scripts/benchmarks``.
-The unified entry points cover RL training with any supported library as well as environment
-stepping without any reinforcement learning library.
-
-Run the benchmark entry points through the Isaac Lab CLI:
-
-.. code-block:: bash
-
-   # benchmark with RSL RL
-   uv run isaaclab benchmark training --rl_library rsl_rl --task=Isaac-Cartpole
-
-   # benchmark with RL Games
-   uv run isaaclab benchmark training --rl_library rl_games --task=Isaac-Cartpole
-
-   # benchmark without RL libraries (environment stepping only)
-   uv run isaaclab benchmark runtime --task=Isaac-Cartpole
+Reproduce These Workloads
+-------------------------
 
-Each benchmark emits a schema-v1 JSON bundle at the end of the run, which includes data on the
-startup times, runtime statistics such as the time taken for each simulation or rendering step,
-as well as overall environment FPS for stepping the environment, performing inference during
-rollout, and training.
+Use the supported runtime, play, training, and startup workflows described in
+:ref:`testing_benchmarks`. Record the task, backend, CPU, GPU, software revision,
+environment count, seed, warm-up, and measured window with every published
+result.