Shiki42/q4a-putcab-sequential-dp-e698-step83405

E698 PutCab Sequential Diffusion Policy

This is the inference checkpoint from CTR Experiment E698-R001, trained with the official LeRobot 0.4.4 Diffusion Policy implementation for the RoboTwin PutCab simulation task. It contains the final step 83,405 pretrained_model files: model weights, resolved configuration, and policy preprocessor and postprocessor state. Training optimizer, RNG, and data-loader state are not included.

Provenance

• Training Experiment: E698-R001, completed with exit code 0 after 83,405 optimizer updates.
• Training data: Shiki42/PutCab-Sequential-Train50-V4, immutable revision c6022cf9164b34c7244dfede8ece4d20a999ab82 (E686 qualification). Sequential left-first and right-first episodes are balanced; IdleMask is disabled for this non-CTR arm.
• LeRobot source commit: 8fff0fde7c79f23a93d845d1a50e985de01f8b8a.
• Training Run receipt SHA-256: c5408db41c68ff601176b83baa8dbc96cbb69351b342696ce5afe6858881e6db.
• E698 independent CPU checkpoint qualification receipt SHA-256: f2e853faf7a42e3528f0a49aab1c141666d6ba17ad420d33338558cf013e771e. It checked the exact final checkpoint and fresh CPU reload, including the saved processors, native eight-action queue, and 100 denoising steps.
• Consumed normalization statistics SHA-256: 8011dbea9f43aea0fd6f99823d1920607060d788441070e6c7d41d151f382ab7.

Intended use

This checkpoint is provided for the registered E721 PutCab simulation evaluation. E721 requires its own frozen-scene, runtime, action-queue, and scoring qualification before a success rate can be reported. Publication is not evaluation or audit approval.

SHA256SUMS lists the exact published inference files. Use a pinned repository revision and verify its hashes before inference.