feat: Add success rate to benchmark results

scripts/benchmarks/benchmark_rlgames.py

@@ -95,11 +95,13 @@
 from scripts.benchmarks.utils import (
     get_backend_type,
     get_preset_string,
+    get_success_rate_log,
     log_app_start_time,
     log_convergence,
     log_python_imports_time,
     log_rl_policy_episode_lengths,
     log_rl_policy_rewards,
+    log_rl_policy_success_rates,
     log_runtime_step_times,
     log_scene_creation_time,
     log_simulation_start_time,
@@ -274,6 +276,9 @@ def main(
         log_runtime_step_times(benchmark, rl_training_times, compute_stats=True)
         log_rl_policy_rewards(benchmark, log_data["rewards/iter"])
         log_rl_policy_episode_lengths(benchmark, log_data["episode_lengths/iter"])
+        success_rates = get_success_rate_log(log_data)
+        if success_rates is not None:
+            log_rl_policy_success_rates(benchmark, success_rates)
         log_convergence(
             benchmark,
             log_data["rewards/iter"],

scripts/benchmarks/benchmark_rsl_rl.py

@@ -97,11 +97,13 @@
 from scripts.benchmarks.utils import (
     get_backend_type,
     get_preset_string,
+    get_success_rate_log,
     log_app_start_time,
     log_convergence,
     log_python_imports_time,
     log_rl_policy_episode_lengths,
     log_rl_policy_rewards,
+    log_rl_policy_success_rates,
     log_runtime_step_times,
     log_scene_creation_time,
     log_simulation_start_time,
@@ -268,6 +270,9 @@ def main(
         log_runtime_step_times(benchmark, rl_training_times, compute_stats=True)
         log_rl_policy_rewards(benchmark, log_data["Train/mean_reward"])
         log_rl_policy_episode_lengths(benchmark, log_data["Train/mean_episode_length"])
+        success_rates = get_success_rate_log(log_data)
+        if success_rates is not None:
+            log_rl_policy_success_rates(benchmark, success_rates)
 
         log_convergence(
             benchmark,

scripts/benchmarks/utils.py

@@ -19,6 +19,7 @@
     os.path.dirname(__file__), "..", "..", "source", "isaaclab_tasks", "test", "benchmarking"
 )
 _CONFIGS_YAML = os.path.join(_BENCHMARKING_DIR, "configs.yaml")
+SUCCESS_RATE_LOG_TAGS = ("Metrics/success_rate", "Episode/Metrics/success_rate")
 
 
 def get_backend_type(cli_backend: str) -> str:
@@ -151,6 +152,23 @@ def log_rl_policy_episode_lengths(benchmark: BaseIsaacLabBenchmark, value: list)
     benchmark.add_measurement("train", measurement=measurement)
 
 
+def log_rl_policy_success_rates(benchmark: BaseIsaacLabBenchmark, value: list):
+    if not value:
+        return
+    measurement = ListMeasurement(name="Success Rates", value=value)
+    benchmark.add_measurement("train", measurement=measurement)
+    # Log the best observed success rate as a scalar for benchmark JSON backends.
+    measurement = SingleMeasurement(name="success_rate", value=max(value), unit="float")
+    benchmark.add_measurement("train", measurement=measurement)
+
+
+def get_success_rate_log(log_data: dict) -> list | None:
+    for tag in SUCCESS_RATE_LOG_TAGS:
+        if tag in log_data:
+            return log_data[tag]
+    return None
+
+
 def check_convergence(
     rewards: list[float],
     threshold: float,

source/isaaclab_tasks/config/extension.toml

@@ -1,7 +1,7 @@
 [package]
 
 # Note: Semantic Versioning is used: https://semver.org/
-version = "1.5.29"
+version = "1.5.30"
 
 # Description
 title = "Isaac Lab Environments"

source/isaaclab_tasks/docs/CHANGELOG.rst

@@ -1,6 +1,16 @@
 Changelog
 ---------
 
+1.5.30 (2026-04-28)
+~~~~~~~~~~~~~~~~~~~
+
+Added
+^^^^^
+
+* Added benchmark extraction for ``Metrics/success_rate`` and survival
+  success logging for direct cartpole camera environments.
+
+
 1.5.29 (2026-04-27)
 ~~~~~~~~~~~~~~~~~~~
 

source/isaaclab_tasks/isaaclab_tasks/direct/cartpole/cartpole_camera_env.py

@@ -151,6 +151,11 @@ def _get_dones(self) -> tuple[torch.Tensor, torch.Tensor]:
     def _reset_idx(self, env_ids: Sequence[int] | None):
         if env_ids is None:
             env_ids = self._cartpole._ALL_INDICES
+
+        # Log survival success rate before resetting.
+        survived = self.reset_time_outs[env_ids].float()
+        self.extras.setdefault("log", {})["Metrics/success_rate"] = survived.mean().item()
+
         super()._reset_idx(env_ids)
 
         joint_pos = self._cartpole.data.default_joint_pos.torch[env_ids]