Upload 5 files
Browse files- LICENSE-NOTE.txt +1 -0
- README.md +23 -0
- policy.json +17 -0
- policy.onnx +3 -0
- upstream-base-card.md +37 -0
LICENSE-NOTE.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
Inherited weights retain their original upstream terms and attribution in upstream-base-card.md. New own calibration or PPO output-layer updates do not assert a new blanket redistribution license. No new trajectory lookup or seed-specific table was added.
|
README.md
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- microduck-arena
|
| 4 |
+
- onnx
|
| 5 |
+
- robotics
|
| 6 |
+
---
|
| 7 |
+
# Roller Explore016 research checkpoint
|
| 8 |
+
|
| 9 |
+
Native CPU physics with MPS PPO output-layer learning, checkpoint 16. The inherited Pollen neural backbone and the prior own N8 feedback controller are credited in upstream-base-card.md. The exported policy uses only the official 61 observations and emits 14 joint actions.
|
| 10 |
+
|
| 11 |
+
Event: `roller-sprint-2m`. Fixed external command: **1.1**. Accessories: rollers. Input `[1,61]`, output `[1,14]`, float32 ONNX, 50 Hz.
|
| 12 |
+
|
| 13 |
+
## Local validation
|
| 14 |
+
|
| 15 |
+
At fixed command 1.1, the local canonical five-start result was 2.741 seconds, 5/5 finishes. New ordinary-start validation finished 128/128; command-stress validation finished 64/64 at commands 1.0 and 1.2. This checkpoint passed the declared 30 ms noninferiority screen against the earlier general baseline. It does not improve the published Roller best of 2.570 seconds.
|
| 16 |
+
|
| 17 |
+
This is a distinct previously unsubmitted research checkpoint for daily-run points, not a personal-best or podium claim. Official Arena acceptance, completion and points are not guaranteed. Local Apple Silicon results may differ from server results. No Tug policy or cup entrant is changed by publication of this repository.
|
| 18 |
+
|
| 19 |
+
Inherited upstream terms and attribution are retained in upstream-base-card.md and LICENSE-NOTE.txt. No new blanket redistribution license is asserted.
|
| 20 |
+
|
| 21 |
+
Model SHA256: `e3ce2bf96f8ce86cfc06c2f452b9caf187e9dc51e1f89c9e4b88c5a87eab3ec0`.
|
| 22 |
+
|
| 23 |
+
Simulation only. No real-robot safety or reliability guarantee.
|
policy.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"event": "roller-sprint-2m",
|
| 3 |
+
"command_vx": 1.1,
|
| 4 |
+
"accessories": [
|
| 5 |
+
"rollers"
|
| 6 |
+
],
|
| 7 |
+
"policy": "policy.onnx",
|
| 8 |
+
"sha256": "e3ce2bf96f8ce86cfc06c2f452b9caf187e9dc51e1f89c9e4b88c5a87eab3ec0",
|
| 9 |
+
"input": [
|
| 10 |
+
1,
|
| 11 |
+
61
|
| 12 |
+
],
|
| 13 |
+
"output": [
|
| 14 |
+
1,
|
| 15 |
+
14
|
| 16 |
+
]
|
| 17 |
+
}
|
policy.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e3ce2bf96f8ce86cfc06c2f452b9caf187e9dc51e1f89c9e4b88c5a87eab3ec0
|
| 3 |
+
size 808526
|
upstream-base-card.md
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- microduck
|
| 4 |
+
- robotics
|
| 5 |
+
- onnx
|
| 6 |
+
- microduck-slot:walk
|
| 7 |
+
library_name: microduck
|
| 8 |
+
pipeline_tag: robotics
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# dfreq-rollers — N8 coupled feedback update
|
| 12 |
+
|
| 13 |
+
Frozen Roller policy at external command **1.1**. Local independent evaluation: **128/128 finishes**, mean **2.7175 s**, median 2.720 s, maximum 2.760 s. The actual official local CLI at command 1.1 returned **2.715 s, 5/5 finishes** on its prescribed five seeds. These are local measurements, not a new Arena server result or leaderboard claim.
|
| 14 |
+
|
| 15 |
+
## Method and contract
|
| 16 |
+
|
| 17 |
+
The original credited Pollen neural backbone remains frozen. N8 adds a memoryless coupled feedback head using 16 bounded projections of existing hidden features plus a bias, with 238 learned coefficients. Episodic derivative-free optimization and accepted-direction search produced the selected model; this is not PPO. Its total added residual is bounded by 0.12 rad. Actions are radian offsets from DEFAULT_POSE; that residual cap is an architectural choice, not an official whole-action limit. The inherited feedback command calibration is retained.
|
| 18 |
+
|
| 19 |
+
The deployed model receives only the official 61 raw feedback inputs and emits 14 float32 actions at 50 Hz. No scoring clock, seed identity, reset fingerprint, privileged simulator velocity/position or per-seed action table enters the policy. Previous-action feedback remains the actual previous output. Official physics, preparation and actuator limits were unchanged.
|
| 20 |
+
|
| 21 |
+
## Frozen evaluation
|
| 22 |
+
|
| 23 |
+
N8 and command 1.1 were frozen before one evaluation on 128 previously untouched paired resets (900001–900128). No alternative policy or command was selected from those results. N8 was faster on every pair against both the earlier private fallback and previous published policy. The private fallback also finished 128/128, mean 2.922617 s: N8's paired mean gain was 205.117 ms (paired bootstrap 95% interval 199.883–210.274 ms). The previous public model finished 128/128, mean 3.178906 s; its 10.755 s recovery outlier affects that mean. Median paired gain against it was 390 ms. All individual results, failures from training and fixed promotion gates were retained privately.
|
| 24 |
+
|
| 25 |
+
The original prescribed five seeds were already exposed regression data. Their CLI replication is not fresh validation. The 128 final starts are limited simulation evidence, not a guarantee on all resets, other physics or hardware. Floating-point contact sensitivity limits portability; claims here concern the exported ONNX, not Python training inference. Later private experiments did not replace this validated N8 model.
|
| 26 |
+
|
| 27 |
+
Runtime pins: Arena source `bbddfe587d300063b1d162281d647740738742f5`, runner v5, Python 3.12.12, NumPy 2.4.1, MuJoCo 3.10.0, ONNX Runtime 1.24.4 on Apple Silicon macOS. Preparation is 25 untimed control calls; distance origin resets afterward while momentum is retained.
|
| 28 |
+
|
| 29 |
+
## Artifact, history and credit
|
| 30 |
+
|
| 31 |
+
`policy.onnx` SHA256: `dc660e8128c3ad2446a0cee9c2f8690ccae3dab681509c57010f91a59d78e797`.
|
| 32 |
+
|
| 33 |
+
Previous public revision: [6f282e1a904146ef800ef011b02863c29e2a420b](https://huggingface.co/djduecaster/dfreq-rollers/tree/6f282e1a904146ef800ef011b02863c29e2a420b), model SHA256 `58122a7287b3330d167f5659a5a8fc9f65c7456f7851e5e5d0a9673ecc505d92`. That previous entry's historical server score was 3.081 s; publication of N8 alone does not submit a new Arena entry.
|
| 34 |
+
|
| 35 |
+
Original neural network: [Pollen Robotics Roller Sprint reference](https://huggingface.co/pollen-robotics/microduck-roller-sprint-2m-reference/tree/c72fb92594e8b415487e87dfeed509141afc5e25), revision `c72fb92594e8b415487e87dfeed509141afc5e25`, SHA256 `01edf63a529a8396d46dac62bb427fda1ce67685611e72ddf790bf46f610c82a`. The manifest training field describes upstream Pollen training. Existing service-scoped remix provenance is preserved. No standalone open-source model license or blanket off-platform redistribution grant is asserted.
|
| 36 |
+
|
| 37 |
+
Simulation only; no real-robot deployment or braking/stability safety claim. This publication changes only the Roller repository, with owner approval.
|