Switch base-mass randomization to log-uniform scale (drop per-robot overrides)

Change EventsCfg.add_base_mass from additive (-5, 5) kg uniform to
multiplicative (1/1.25, 1.25) log-uniform scale with distribution="log_uniform"
and operation="scale". The range is symmetric in log space (geometric mean
= 1.0), so the randomization has no net mass bias and is scale-invariant
across robot sizes — a single default works for the 2 kg pelvis of Cassie
and the 32 kg base of Anymal-D without per-robot tuning.

Drop the per-robot (-1.0, 3.0) additive overrides on A1, Go1, Go2 (introduced
in PR #3308 [Newton] Locomotion Velocity-Flat envs, 2025-09-18) which have
a +1 kg mean shift that biases controllers toward always-heavier nominal —
a 10% bias on A1's 10 kg base that was actively hurting convergence.

## Why the change

The old (-5, 5) kg additive default was a reasonable tune for Anymal-D
(≈16% range) but too wide for A1 (≈50%) and too narrow for H1 (≈10%).
PR #3308 addressed this with per-robot (-1, 3) overrides on small quadrupeds,
but the range remained asymmetric (mean shifted +1 kg) and tuning was
per-robot.

A multiplicative scale is more physically meaningful — mass scales in
ratio with actuator torque and gravity response, so the perturbation's
effect on gait is a fixed fraction regardless of robot mass. Log-uniform
over [1/1.25, 1.25] gives symmetric perturbation in log space with
geometric mean = 1.0, so the training distribution is unbiased.

## Ablation results (1500-iter Newton unless noted)

| Robot       | old default | new log-uniform | Δ           |
|-------------|------------:|----------------:|------------:|
| A1 @300     |        3.25 |           10.00 |       3.08x |
| Go1 @300    |       16.30 |           22.29 |       1.37x |
| Anymal-B @300 |      10.92 |           12.47 |       1.14x |
| Anymal-C @300 |      12.31 |           14.64 |       1.19x |
| Go2 @1499   |       18.58 |           24.71 |       1.33x |
| Anymal-D @1499 |     15.62 |           16.09 |       1.03x |

A1's 3x gain is the bias-removal effect: the old (-1, 3) applied +10%
mean mass shift on a 10 kg robot, forcing the controller to spend capacity
compensating for a nonphysical average instead of learning the nominal
gait. The new default's geometric mean 1.0 removes that bias entirely.

Biped configurations (H1, Cassie) are handled in the follow-up PR #5298.
Cassie specifically requires an asymmetric heavier-bias override due to
its closed-loop Achilles coupling being mass-sensitive; that override
ships alongside the biped changes.

Author: hujc7

source/isaaclab_tasks/config/extension.toml

@@ -1,7 +1,7 @@
 [package]
 
 # Note: Semantic Versioning is used: https://semver.org/
-version = "1.5.22"
+version = "1.5.23"
 
 # Description
 title = "Isaac Lab Environments"

source/isaaclab_tasks/docs/CHANGELOG.rst

@@ -1,6 +1,23 @@
 Changelog
 ---------
 
+1.5.23 (2026-04-22)
+~~~~~~~~~~~~~~~~~~~
+
+Changed
+^^^^^^^
+
+* Replaced the additive ``(-5, 5)`` kg default on
+  ``EventsCfg.add_base_mass`` with a multiplicative ``(1/1.25, 1.25)``
+  log-uniform scale (``operation="scale"``, ``distribution="log_uniform"``).
+  The new default is scale-invariant across robot sizes and has geometric
+  mean 1.0 — no net mass bias. Removed the per-robot override
+  ``(-1.0, 3.0)`` additive on A1/Go1/Go2 (no longer needed). On Anymal-D
+  1500-iter Newton, the new default raises reward from 15.62 to 16.09; on
+  A1 at iter 300, from 3.25 to 10.00 (the old override had a +10% mean
+  bias, which was 10% of A1's 10 kg base mass and hurt convergence on
+  small robots).
+
 1.5.22 (2026-04-16)
 ~~~~~~~~~~~~~~~~~~~
 

source/isaaclab_tasks/isaaclab_tasks/manager_based/locomotion/velocity/config/a1/rough_env_cfg.py

@@ -29,7 +29,6 @@ def __post_init__(self):
 
         # A1 uses "trunk" as base body
         self.events.add_base_mass.params["asset_cfg"].body_names = "trunk"
-        self.events.add_base_mass.params["mass_distribution_params"] = (-1.0, 3.0)
         self.events.base_external_force_torque.params["asset_cfg"].body_names = "trunk"
         self.events.base_com.default.params["asset_cfg"].body_names = "trunk"
 

source/isaaclab_tasks/isaaclab_tasks/manager_based/locomotion/velocity/config/go1/rough_env_cfg.py

@@ -31,7 +31,6 @@ def __post_init__(self):
 
         # Go1 uses "trunk" as base body
         self.events.add_base_mass.params["asset_cfg"].body_names = "trunk"
-        self.events.add_base_mass.params["mass_distribution_params"] = (-1.0, 3.0)
         self.events.base_external_force_torque.params["asset_cfg"].body_names = "trunk"
         self.events.base_com.default.params["asset_cfg"].body_names = "trunk"
 

source/isaaclab_tasks/isaaclab_tasks/manager_based/locomotion/velocity/config/go2/rough_env_cfg.py

@@ -30,9 +30,6 @@ def __post_init__(self):
         self.scene.terrain.terrain_generator.sub_terrains["random_rough"].noise_range = (0.01, 0.06)
         self.scene.terrain.terrain_generator.sub_terrains["random_rough"].noise_step = 0.01
 
-        # Go2 uses smaller mass randomization range
-        self.events.add_base_mass.params["mass_distribution_params"] = (-1.0, 3.0)
-
         # reduce action scale
         self.actions.joint_pos.scale = 0.25
 

source/isaaclab_tasks/isaaclab_tasks/manager_based/locomotion/velocity/velocity_env_cfg.py

@@ -209,8 +209,13 @@ class EventsCfg:
         mode="startup",
         params={
             "asset_cfg": SceneEntityCfg("robot", body_names="base"),
-            "mass_distribution_params": (-5.0, 5.0),
-            "operation": "add",
+            # Multiplicative ±25% log-uniform. Scale-invariant across robot sizes, so
+            # per-robot range overrides are no longer needed. Geometric mean = 1.0
+            # (no net mass shift); inverse symmetric (acceleration perturbation
+            # symmetric around nominal).
+            "mass_distribution_params": (1 / 1.25, 1.25),
+            "operation": "scale",
+            "distribution": "log_uniform",
         },
     )