|
Download docs/vector-extras.md from ccnets/causal-gpt-rl-unity-envs: direct link, hf CLI and curl.
- Browser
- Download file 4.59 kB
-
https://huggingface.co/datasets/ccnets/causal-gpt-rl-unity-envs/resolve/main/docs/vector-extras.md
- Command line
-
hf download hf://datasets/ccnets/causal-gpt-rl-unity-envs/docs/vector-extras.md
-
curl -L -o vector-extras.md https://huggingface.co/datasets/ccnets/causal-gpt-rl-unity-envs/resolve/main/docs/vector-extras.md
4.59 kB
| # Four vector-only simulation additions | |
| Match3VectorObs, Basic, 3DBall, and PushBlockWithInput were added on 2026-10-22. | |
| Windows and Linux x86-64 players were built with Unity 6000.0.40f1 from the local | |
| ML-Agents release-23 project. Model references were removed from scene copies; | |
| BuildReport.packedAssets contains no ONNX/NN policies. Every live observation | |
| sensor was verified to be rank one with the Windows ML-Agents communicator. | |
| The Match3 scene includes only the VectorObs agents; VisualObs and heuristic | |
| boards were removed. No camera rendering is needed to control any of these four. | |
| ## Running | |
| Use Python 3.10 and `tools/requirements-simulation.txt`. Download the desired | |
| Windows folder (or `linux/<Name>/**`), the separate root `<Name>.onnx`, and | |
| `tools/probe_vector_extras.py`: | |
| ```python | |
| from huggingface_hub import snapshot_download | |
| snapshot_download( | |
| repo_id="ccnets/causal-gpt-rl-unity-envs", repo_type="dataset", | |
| allow_patterns=["Match3VectorObs/**", "Match3VectorObs.onnx", "tools/**"], | |
| local_dir="unity-envs", | |
| ) | |
| ``` | |
| ```bash | |
| python unity-envs/tools/probe_vector_extras.py --build unity-envs/Match3VectorObs/Match3VectorObs.exe --policies unity-envs --report probe.json --ticks 1000 | |
| ``` | |
| Use `--random` to drive without any policy, or `--graphics --time-scale 1` for a | |
| visible window. Each concurrent player needs a distinct `--worker-id`. Linux | |
| players require execute permission (`chmod +x`) and use the `.x86_64` executable. | |
| These are simulations and validation tools; no trajectories or quality tiers | |
| are generated here. | |
| ## Contracts and caveats | |
| - Match3VectorObs represents ordinary and special tiles as separate flat vectors. | |
| Respect the dynamic action mask: each action selects an adjacent cell swap. | |
| Boards vary in size within the configured maxima. Episodes interrupt after | |
| 500 moves; an interruption is a time limit, not task failure. | |
| - Basic observes its current position as a one-hot vector of length 20. Its | |
| three actions move left, stay, or move right. The small goal rewards +0.1, | |
| the larger goal +1, and each step costs 0.01. Episode reset reloads its scene. | |
| - 3DBall observes platform rotation, relative ball position, and ball velocity. | |
| Two continuous actions tilt the platform. This differs from 3DBallHard, | |
| which hides velocity and uses stacked observations. | |
| - PushBlockWithInput uses Unity Input System actuators: continuous movement | |
| and rotation plus a binary jump action. Its behavior is still named | |
| `PushBlock?team=0`. Do not reuse the older discrete `PushBlock.onnx`: the | |
| separate hybrid policy is published as `PushBlockWithInput.onnx`. This is a | |
| filename-only rename of `PushBlockWithInput/TFModels/PushBlock.onnx`. | |
| Its legacy ONNX takes a concatenated vector; the included runner joins the | |
| live ray sensors in their ML-Agents order and samples masked discrete logits. | |
| ## Validation and reproduction | |
| Each Windows player completed 1,000 external-stock-policy ML-Agents steps, | |
| with finite observations/actions/rewards, accepted decisions, and episode | |
| boundaries. Each Linux player passed a native Ubuntu/WSL engine startup check; | |
| Linux policy rollouts were not tested. Reports are under | |
| `validation/vector-extras/`. Reward sums are raw smoke-test results, not tier | |
| calibration or comparable task scores. | |
| `tools/VectorExtrasBuild.cs` stages scene copies, clears serialized model | |
| references, sets BehaviorType.Default, and rejects packed ONNX/NN assets. | |
| Install it in `Assets/Editor/` in a separate release-23 project. Invoke Unity: | |
| ```text | |
| -batchmode -force-d3d11 -projectPath <project> -buildTarget Win64 -executeMethod VectorExtrasBuild.Build -simulationOutput <output> | |
| ``` | |
| For Linux replace `-buildTarget Win64` with `-buildTarget Linux64` and add | |
| `-simulationTarget StandaloneLinux64`. | |
| The source project and its packages must be present locally; this helper is | |
| not a complete source archive. Published manifest action specs include native | |
| ActuatorComponent contributions and are cross-checked against live probes. | |
| ## Verified policy provenance | |
| All four ONNX files are byte-identical to their release-23 source files: | |
| - `Match3VectorObs.onnx`: SHA-256 `876995464fa883b992b493f8daa7a73179d04e5097b8d26468ff688a31356bf9`. | |
| - `Basic.onnx`: SHA-256 `76685fb46f3618dcdc1aa870d36a541f6550cdb2ce639232547d195aa6f01a2f`. | |
| - `3DBall.onnx`: SHA-256 `496dc01d14506b635268fc62ff0d58a16be71fe3a83e47c1ac72984ce69a1864`. | |
| - `PushBlockWithInput.onnx`: SHA-256 `270d26f24ab34fa3238d7503b89b08b605062649b12fdd63d22f0556d32be5a3`. | |