Replace WanOFT latency PPO with accepted 10 FPS policy

Selected at 23,040,000 steps from W&B run s463nqus; independent success_at_end 240/256 (93.75%).

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-wanoft-rtx3090-v1/README.md

@@ -19,17 +19,17 @@ temporal latency profile measured for WanOFT on one RTX 3090. The policy uses
 - Environment: `InterceptGrabFast-VLA-v0`, `normalized_dense` reward, 60-step episodes.
 - Observation: 49D privileged state; actor and critic each use three 256-unit tanh layers.
 - Action: 7D `pd_ee_delta_pose`, one action per accepted request.
-- Control frequency: 20 Hz; physics frequency: 100 Hz.
+- Benchmark and observation clocks: 10 FPS; native control: 20 Hz; physics: 100 Hz.
 - Seed: 123; 1,024 environments; 60-step rollouts; 32 minibatches; eight update epochs.
 - Learning rate: `1e-4`; PPO target KL: `0.2`; gamma: `0.99`; GAE lambda: `0.95`.
 - Latency: `temporal`, training latency seed 0, one simulated worker,
   `latest_ready` ordering, and `hold` between action arrivals.
 - The profile's mean observation-to-action latency is 300.54 ms. The joint model
   represents both worker occupancy and action-ready time.
-- Selected checkpoint: 69,120,000 environment steps, iteration 1125, chosen by
-  the highest periodic `eval/success_at_end` during the 150M-step run.
+- Selected checkpoint: 23,040,000 environment steps, iteration 375, chosen by
+  the highest periodic `eval/success_at_end`; training stopped at 31,150,080 steps.
 
-Training run: [sw3b981e](https://wandb.ai/saberrr-zju/MIKASA-Robo-dataset-collectors/runs/sw3b981e).
+Training run: [s463nqus](https://wandb.ai/saberrr-zju/MIKASA-Robo-dataset-collectors/runs/s463nqus).
 
 The exact latency profile is
 [`profiles/wanoft/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c`](https://huggingface.co/datasets/latency-sensitive-bench/profiles/tree/66414463f2a609316da63935941a79abf0a73ce9/profiles/wanoft/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c),
@@ -37,17 +37,17 @@ pinned to revision `66414463f2a609316da63935941a79abf0a73ce9`.
 
 ## Independent evaluation
 
-The evaluation used 64 parallel environments, 240 control steps per environment
+The evaluation used 64 parallel environments, 240 benchmark steps per environment
 (four complete episodes each), environment seed `20260910`, and independent
 latency seed `271828`. Actions were deterministic policy means.
 
 | Metric | Result |
 | --- | --- |
-| Success at least once | 161 / 256 = **62.89%** |
-| Success at the end | 122 / 256 = **47.66%** |
-| Mean episode return | 22.5840 |
-| Mean episode length | 60 control steps |
-| Accepted / capacity-dropped requests | 2,560 / 12,800 |
+| Success at least once | 252 / 256 = **98.44%** |
+| Success at the end | 240 / 256 = **93.75%** |
+| Mean episode return | 30.8664 |
+| Mean episode length | 60 benchmark steps |
+| Accepted / capacity-dropped requests | 5,119 / 10,241 |
 
 ## Download and evaluate
 

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-wanoft-rtx3090-v1/checkpoint.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ec32a1d150f295d7f27d8d2dabdda34966b91389ff3d6568c09a75da98334ea
+oid sha256:ba0f485b8dec4da11836eb64f6863df172744b0922e654ace644dd4668d82e4b
 size 1169937

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-wanoft-rtx3090-v1/eval.yaml

@@ -1,6 +1,6 @@
 env:
-  env_fps: 20
-  obs_fps: 20
+  env_fps: 10
+  obs_fps: 10
   noop_action: [0, 0, 0, 0, 0, 0, 0]
 executor:
   simulated_worker_capacity: 1
@@ -16,7 +16,7 @@ latency:
   seed: 271828
   profile_path: profiles/wanoft/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c/profile.json
 training:
-  exp_name: small-ppo-mikasa-intercept-grab-fast-wanoft-rtx3090-temporal-scratch
+  exp_name: small-ppo-mikasa-intercept-grab-fast-wanoft-rtx3090-temporal-clock10-scratch
   checkpoint: checkpoint.pt
   resume: null
   seed: 20260910

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-wanoft-rtx3090-v1/provenance.json

@@ -2,12 +2,12 @@
   "task": "mikasa-intercept-grab-fast",
   "model": "ppo-mlp-state",
   "status": "published",
-  "run_id": "sw3b981e",
-  "wandb_url": "https://wandb.ai/saberrr-zju/MIKASA-Robo-dataset-collectors/runs/sw3b981e",
+  "run_id": "s463nqus",
+  "wandb_url": "https://wandb.ai/saberrr-zju/MIKASA-Robo-dataset-collectors/runs/s463nqus",
   "checkpoint": "checkpoint.pt",
   "checkpoint_bytes": 1169937,
-  "checkpoint_global_step": 69120000,
-  "checkpoint_iteration": 1125,
+  "checkpoint_global_step": 23040000,
+  "checkpoint_iteration": 375,
   "selection_metric": "eval/success_at_end",
   "initialization": "from_scratch",
   "training_seed": 123,
@@ -17,7 +17,7 @@
     "path": "profiles/wanoft/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c"
   },
   "trainer_source": {
-    "wandb_run_id": "sw3b981e",
+    "wandb_run_id": "s463nqus",
     "path": "code/third_party/MIKASA-Robo/mikasa_robo_suite/vla/dataset_collectors/get_dataset_collectors_ckpt.py"
   },
   "evaluation": "strict-eval.json"

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-wanoft-rtx3090-v1/strict-eval.json

@@ -1,21 +1,21 @@
 {
   "episodes": 256,
   "metrics": {
-    "success_once": 0.62890625,
-    "return": 22.58401107788086,
+    "success_once": 0.984375,
+    "return": 30.86635398864746,
     "episode_len": 60.0,
-    "reward": 0.3764001727104187,
-    "success_at_end": 0.4765625
+    "reward": 0.5144392251968384,
+    "success_at_end": 0.9375
   },
-  "admitted_count": 2560,
-  "dropped_count": 12800,
+  "admitted_count": 5119,
+  "dropped_count": 10241,
   "checkpoint": "checkpoint.pt",
-  "checkpoint_global_step": 69120000,
+  "checkpoint_global_step": 23040000,
   "evaluation_config": "eval.yaml",
-  "success_rate": 0.4765625,
+  "success_rate": 0.9375,
   "success_metric": "success_at_end",
-  "success_once_count": 161,
-  "success_at_end_count": 122,
+  "success_once_count": 252,
+  "success_at_end_count": 240,
   "environment_seed": 20260910,
   "latency_seed": 271828,
   "deterministic": true

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-wanoft-rtx3090-v1/train.yaml

@@ -1,6 +1,6 @@
 env:
-  env_fps: 20
-  obs_fps: 20
+  env_fps: 10
+  obs_fps: 10
   noop_action: [0, 0, 0, 0, 0, 0, 0]
 executor:
   simulated_worker_capacity: 1
@@ -16,7 +16,7 @@ latency:
   seed: 0
   profile_path: profiles/wanoft/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c/profile.json
 training:
-  exp_name: small-ppo-mikasa-intercept-grab-fast-wanoft-rtx3090-temporal-scratch
+  exp_name: small-ppo-mikasa-intercept-grab-fast-wanoft-rtx3090-temporal-clock10-scratch
   checkpoint: null
   resume: null
   seed: 123