Dual-shepherd soft axis-split (HERDING_AXIS_LEAK)

The strict 100/0 axis mask reaches drive standoff and deadlocks
because each dog has only one degree of freedom left to push the
flock. Soften the mask: each dog leads its assigned axis (full gain)
and contributes ``HERDING_AXIS_LEAK`` on the other axis. ``0.0`` is
the old strict behaviour; ``1.0`` is no mask (both dogs run full
policy, role-redundant). Default ``0.3`` breaks the deadlock while
preserving the "one dog per axis" coordination story.

Implementation:
* `controllers/shepherd_dog/shepherd_dog.py` reads
  `HERDING_AXIS_LEAK` from env / runtime cfg (clamped to [0, 1]),
  prints it next to the axis tag, and multiplies the off-axis
  velocity component by it instead of zeroing.
* `tools/run_webots.sh` writes `HERDING_AXIS_LEAK` into
  `herding_runtime.cfg` so Webots-stripped controller subprocesses
  still see it; defaults to 0.3 when unset.

Webots smoke test (HERDING_NDOGS=2, HERDING_AXIS_LEAK=0.3, strombom,
diff/field, 5 sheep, LiDAR perception, no GT): **5/5 penned at step
13204**, vs the strict 100/0 mask which timed out at 0/5. Penning
trail 1/5 → 2/5 → 4/5 → 5/5 between steps 6200 and 13400 — slower
than single-dog (Strömbom diff/field n=5: 7528) as expected since
the work is split, but the coordination demonstrably succeeds.

This gives the writeup a clean three-row ablation:
  α=0.0  (strict)  → deadlock, 0/5
  α=0.3  (default) → 5/5 @ 13204
  α=1.0  (no mask) → both dogs run full policy (single-dog
                     baseline applied twice; no axis story)

126 pytest cases still pass.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

controllers/shepherd_dog/shepherd_dog.py

@@ -269,9 +269,19 @@ timestep = int(robot.getBasicTimeStep())
 
 # Multi-shepherd axis split. When the launcher creates two dog instances
 # it sets each robot's customData to "axis=x" or "axis=y"; the controller
-# then masks the off-axis component of every action so the two dogs
+# then attenuates the off-axis component of every action so the two
-# share the herding workload along orthogonal axes (one closes flank
+# dogs share the herding workload along orthogonal axes. customData
-# distance in x, the other in y). customData empty = single-dog mode.
+# empty = single-dog mode (no masking).
+#
+# HERDING_AXIS_LEAK controls how strict the mask is:
+#   0.0 → hard mask (off-axis component zeroed; pure axis-split)
+#   1.0 → no mask (both dogs run full action; equivalent to NDOGS=2
+#                  without axis assignment)
+# Defaults to 0.3 — empirically the 100/0 strict mask deadlocks once
+# both dogs reach their drive standoff (the Strömbom target shrinks
+# and each dog has only one degree of freedom). A small leak keeps
+# pressure on the flock while preserving the "one dog leads each
+# axis" coordination story.
 _AXIS_TAG = (robot.getCustomData() or "").strip().lower()
 if _AXIS_TAG.startswith("axis="):
     DOG_AXIS = _AXIS_TAG[5:]
@@ -280,8 +290,14 @@ if _AXIS_TAG.startswith("axis="):
         DOG_AXIS = None
 else:
     DOG_AXIS = None
+try:
+    AXIS_LEAK = float(os.environ.get("HERDING_AXIS_LEAK")
+                      or _runtime_cfg.get("HERDING_AXIS_LEAK", "0.3"))
+except ValueError:
+    AXIS_LEAK = 0.3
+AXIS_LEAK = max(0.0, min(1.0, AXIS_LEAK))
 DOG_NAME = robot.getName()
-print(f"[dog] name={DOG_NAME}  axis={DOG_AXIS}")
+print(f"[dog] name={DOG_NAME}  axis={DOG_AXIS}  leak={AXIS_LEAK:.2f}")
 
 if DRIVE_MODE == "mecanum":
     fl_motor = robot.getDevice("front left wheel motor")
@@ -483,12 +499,14 @@ while robot.step(timestep) != -1:
     # Near-sheep speed modulation (shared by every mode).
     vx, vy = modulate_speed(vx, vy, dog_xy, sheep_positions)
 
-    # Axis-split mask for the dual-dog setup: this dog only commits to
+    # Axis-split mask for the dual-dog setup: this dog leads its
-    # its assigned axis (x or y) so its partner covers the other.
+    # assigned axis (full gain) and contributes AXIS_LEAK on the
+    # off-axis. With LEAK=0 the mask is strict; with LEAK=1 the dogs
+    # run identical full-power policies.
     if DOG_AXIS == "x":
-        vy = 0.0
+        vy *= AXIS_LEAK
     elif DOG_AXIS == "y":
-        vx = 0.0
+        vx *= AXIS_LEAK
 
     # EMA smoothing — kills frame-to-frame action jitter.
     if DRIVE_MODE == "mecanum":

tools/run_webots.sh

@@ -230,6 +230,7 @@ HERDING_DRIVE=$DRIVE
 HERDING_WORLD=$WORLD
 HERDING_LIDAR=$LIDAR_VARIANT
 HERDING_NDOGS=$NDOGS
+HERDING_AXIS_LEAK=${HERDING_AXIS_LEAK:-0.3}
 HERDING_USE_GT=${HERDING_USE_GT:-0}
 EOF