Publish MIKASA PPO trained with QwenPI v3 RTX 3090 latency profile

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-qwenpi_v3-rtx3090-v1/README.md

@@ -0,0 +1,81 @@
+# MIKASA InterceptGrabFast — PPO with QwenPI v3 RTX 3090 latency
+
+A small privileged-state PPO policy trained from random initialization with the
+temporal latency profile measured for QwenPI v3 on one RTX 3090. The policy uses
+49D state observations and deterministic mean actions for evaluation.
+
+## Files
+
+| File | Contents |
+| --- | --- |
+| `checkpoint.pt` | Inference-ready native MIKASA `AgentStateOnly` state dictionary. |
+| `strict-eval.json` | Independent 256-episode evaluation under the temporal profile. |
+| `train.yaml` | Portable scratch-training configuration. |
+| `eval.yaml` | Portable configuration reproducing the independent evaluation. |
+| `provenance.json` | Selected training step, W&B run, and pinned profile source. |
+
+## Training and latency
+
+- 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.
+- 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 148.69 ms. The joint model
+  represents both worker occupancy and action-ready time.
+- Selected checkpoint: 15,360,000 environment steps, iteration 250, chosen by
+  periodic `eval/success_at_end` during the 25M-step run.
+
+Training run: [u46f6pzq](https://wandb.ai/saberrr-zju/MIKASA-Robo-dataset-collectors/runs/u46f6pzq).
+
+The exact latency profile is
+[`profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c`](https://huggingface.co/datasets/latency-sensitive-bench/profiles/tree/66414463f2a609316da63935941a79abf0a73ce9/profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c),
+pinned to revision `66414463f2a609316da63935941a79abf0a73ce9`.
+
+## Independent evaluation
+
+The evaluation used 64 parallel environments, 240 control 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 | 253 / 256 = **98.83%** |
+| Success at the end | 248 / 256 = **96.88%** |
+| Mean episode return | 28.6463 |
+| Mean episode length | 60 control steps |
+| Accepted / capacity-dropped requests | 5,099 / 10,261 |
+
+`success_at_end` is the primary metric. This checkpoint passes the 90%
+acceptance target.
+
+## Download and evaluate
+
+Run from a benchmark checkout with its MIKASA environment installed:
+
+```bash
+BENCH_ROOT="$PWD"
+MODEL_PATH="latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-qwenpi_v3-rtx3090-v1"
+PROFILE_PATH="profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c"
+BUNDLE="$BENCH_ROOT/outputs/published-models/$MODEL_PATH"
+
+hf download latency-sensitive-bench/benchmark-models \
+  --include "$MODEL_PATH/*" --local-dir "$BENCH_ROOT/outputs/published-models"
+hf download latency-sensitive-bench/profiles --repo-type dataset \
+  --revision 66414463f2a609316da63935941a79abf0a73ce9 \
+  --include "$PROFILE_PATH/*" --local-dir "$BUNDLE"
+
+cd "$BUNDLE"
+CUDA_VISIBLE_DEVICES=0 \
+PYTHONPATH="$BENCH_ROOT:$BENCH_ROOT/third_party/MIKASA-Robo" \
+  "$BENCH_ROOT/third_party/MIKASA-Robo/.venv/bin/python" \
+  "$BENCH_ROOT/third_party/MIKASA-Robo/mikasa_robo_suite/vla/dataset_collectors/get_dataset_collectors_ckpt.py" \
+  --env-id InterceptGrabFast-VLA-v0 \
+  --latency-config eval.yaml --latency-output-dir evaluation
+```
+
+For scratch-training reproduction, use `train.yaml` in the same command and
+choose a separate latency output directory.

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-qwenpi_v3-rtx3090-v1/checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4ac89c11746e7d3d6c890ef97b37fd6bd6c091a55d2af38dd5cfaa7d0e3f7e1
+size 1169937

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-qwenpi_v3-rtx3090-v1/eval.yaml

@@ -0,0 +1,43 @@
+env:
+  env_fps: 20
+  obs_fps: 20
+  noop_action: [0, 0, 0, 0, 0, 0, 0]
+executor:
+  simulated_worker_capacity: 1
+scheduler:
+  ordering_policy: latest_ready
+  hold_policy: hold
+profile_source:
+  repo_id: latency-sensitive-bench/profiles
+  revision: 66414463f2a609316da63935941a79abf0a73ce9
+  path: profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c
+latency:
+  method: temporal
+  seed: 271828
+  profile_path: profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c/profile.json
+training:
+  exp_name: small-ppo-mikasa-intercept-grab-fast-qwenpi-v3-rtx3090-temporal-scratch
+  checkpoint: checkpoint.pt
+  resume: null
+  seed: 20260910
+  num_envs: 1
+  num_eval_envs: 64
+  num_steps: 60
+  num_eval_steps: 240
+  num_minibatches: 32
+  update_epochs: 8
+  learning_rate: 0.0001
+  target_kl: 0.2
+  gamma: 0.99
+  gae_lambda: 0.95
+  clip_coef: 0.2
+  ent_coef: 0.0
+  vf_coef: 0.5
+  max_grad_norm: 0.5
+  anneal_lr: false
+  finite_horizon_gae: true
+  total_timesteps: 60
+  eval_freq: 25
+  track: false
+  evaluate: true
+  save_model: false

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-qwenpi_v3-rtx3090-v1/provenance.json

@@ -0,0 +1,24 @@
+{
+  "task": "mikasa-intercept-grab-fast",
+  "model": "ppo-mlp-state",
+  "status": "accepted",
+  "run_id": "u46f6pzq",
+  "wandb_url": "https://wandb.ai/saberrr-zju/MIKASA-Robo-dataset-collectors/runs/u46f6pzq",
+  "checkpoint": "checkpoint.pt",
+  "checkpoint_bytes": 1169937,
+  "checkpoint_global_step": 15360000,
+  "checkpoint_iteration": 250,
+  "selection_metric": "eval/success_at_end",
+  "initialization": "from_scratch",
+  "training_seed": 123,
+  "profile": {
+    "repo_id": "latency-sensitive-bench/profiles",
+    "revision": "66414463f2a609316da63935941a79abf0a73ce9",
+    "path": "profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c"
+  },
+  "trainer_source": {
+    "wandb_run_id": "u46f6pzq",
+    "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-qwenpi_v3-rtx3090-v1/strict-eval.json

@@ -0,0 +1,22 @@
+{
+  "episodes": 256,
+  "metrics": {
+    "success_once": 0.98828125,
+    "return": 28.646331787109375,
+    "episode_len": 60.0,
+    "reward": 0.47743886709213257,
+    "success_at_end": 0.96875
+  },
+  "admitted_count": 5099,
+  "dropped_count": 10261,
+  "checkpoint": "checkpoint.pt",
+  "checkpoint_global_step": 15360000,
+  "evaluation_config": "eval.yaml",
+  "success_rate": 0.96875,
+  "success_metric": "success_at_end",
+  "success_once_count": 253,
+  "success_at_end_count": 248,
+  "environment_seed": 20260910,
+  "latency_seed": 271828,
+  "deterministic": true
+}

latency-aware/mikasa-intercept-grab-fast/small-policy/ppo-qwenpi_v3-rtx3090-v1/train.yaml

@@ -0,0 +1,41 @@
+env:
+  env_fps: 20
+  obs_fps: 20
+  noop_action: [0, 0, 0, 0, 0, 0, 0]
+executor:
+  simulated_worker_capacity: 1
+scheduler:
+  ordering_policy: latest_ready
+  hold_policy: hold
+profile_source:
+  repo_id: latency-sensitive-bench/profiles
+  revision: 66414463f2a609316da63935941a79abf0a73ce9
+  path: profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c
+latency:
+  method: temporal
+  seed: 0
+  profile_path: profiles/qwenpi_v3/1x-rtx3090/mikasa_intercept_grab_fast/instance_3a0d42681a03715c/profile.json
+training:
+  exp_name: small-ppo-mikasa-intercept-grab-fast-qwenpi-v3-rtx3090-temporal-scratch
+  checkpoint: null
+  resume: null
+  seed: 123
+  num_envs: 1024
+  num_eval_envs: 16
+  num_steps: 60
+  num_eval_steps: 60
+  num_minibatches: 32
+  update_epochs: 8
+  learning_rate: 0.0001
+  target_kl: 0.2
+  gamma: 0.99
+  gae_lambda: 0.95
+  clip_coef: 0.2
+  ent_coef: 0.0
+  vf_coef: 0.5
+  max_grad_norm: 0.5
+  anneal_lr: false
+  finite_horizon_gae: true
+  total_timesteps: 25000000
+  eval_freq: 25
+  track: true