Default GAIT to snow, so the snow gait runs without setting a Space variable

README.md

@@ -41,10 +41,10 @@ crop (`rollout.py`; MJX + `MUJOCO_GL=egl`) and reports how long it stayed up and
 how far it walked. `NUM_EVALS` (default 40) sets how many evals — and therefore
 log lines and checkpoint uploads — a run makes.
 
-## Snow gait (`GAIT=snow`)
+## Snow gait (`GAIT=snow`, the default)
 
 Playground's joystick task is tuned for pavement: a 15 cm swing-height reference
-at 1.25–1.5 Hz with the hips pinned to a narrow stance. `GAIT=snow` retunes it
+at 1.25–1.5 Hz with the hips pinned to a narrow stance. This Space defaults to `GAIT=snow`, which retunes it
 for deep snow via `snow_gait.py` — a 22 cm swing (`FOOT_HEIGHT`), a slower
 1.0–1.3 Hz cadence (`GAIT_FREQ`) with more air-time reward for longer strides, a
 relaxed hip-deviation cost so the stance can widen, and a new cost for letting

app.py

@@ -40,7 +40,11 @@ SNOW_DEPTH = tuple(float(v) for v in os.environ.get("SNOW_DEPTH", "0.0,0.08").sp
 TASK = os.environ.get("TASK", "walk").strip().lower()
 # "snow" retunes the walking gait for deep snow: higher swing, wider stance,
 # slower cadence, and a lateral foot-separation cost (see snow_gait.py).
-GAIT = os.environ.get("GAIT", "street").strip().lower()
+# Defaults to the snow gait: this Space trains a G1 for snow, and the street
+# gait is what the 200M-step runs plateaued on (the policy crossed its own feet
+# after ~45 steps and took the -100 termination every episode). Set GAIT=street
+# for Playground's stock pavement tuning.
+GAIT = os.environ.get("GAIT", "snow").strip().lower()
 FOOT_HEIGHT = float(os.environ.get("FOOT_HEIGHT", 0.22))        # swing reference, m
 FOOT_SEPARATION = float(os.environ.get("FOOT_SEPARATION", 0.20))  # min lateral gap, m
 GAIT_FREQ = tuple(float(v) for v in os.environ.get("GAIT_FREQ", "1.0,1.3").split(","))

himalaya_terrain.py

@@ -36,7 +36,12 @@ def sample_patch(dem, mpp, rng, patch_m, res, arena, relief,
     n = dem.shape[0]
     px = patch_m / mpp
     margin = px * 0.75
-    ci, cj = rng.uniform(margin, n - margin, size=2)
+    # A patch approaching the size of the DEM leaves no room to move the crop
+    # around; fall back to the centre rather than sampling an empty range.
+    if n - margin <= margin:
+        ci = cj = (n - 1) / 2.0
+    else:
+        ci, cj = rng.uniform(margin, n - margin, size=2)
     theta = rng.uniform(0, 2 * np.pi)
 
     u = (np.arange(res) - (res - 1) / 2) / (res - 1) * px