Search is not available for this dataset
video video |
|---|
End of preview. Expand in Data Studio
KinDER adaptive-reasoning: evaluation results
Outputs of the KinDER experiments in GitHub RobertCHU01/adaptive-reasoning
(read project_knowledge/ there for what each run was and what it found). Videos,
per-episode traces and HTML reports; no model weights.
| Path | Content |
|---|---|
pi05_evals/<run>/<step>/eval_<job>/ |
pi05 simulation evals (KinDER test protocol, seed 301 unless noted). <run>: kinder2task_b64_1gpu_v1 (pi05_kinder_official), kinder2task_rel_b64_v1 (pi05_kinder_rel), kinder_subgoal_eef_b64_v1 (subgoal-conditioned pi05_kinder_subgoal_eef, with GT / oracle / VLM subgoal sources). Each has <task>/episode_XXX_{success,failure}.mp4, index.html, results.json/traces. Job ids match the notes in project_knowledge/ |
wm_best_of_n/ |
Best-of-N action selection runs (pi05 samples, WAP imagines, a VLM picks), project_knowledge/wm_best_of_n.md |
wm_ckpt_tests/ |
test_wm_ckpt.py results (WAP prediction PSNR vs copy-last on held-out pi05 rollouts) |
point_bench/ |
Offline VLM 3D-point benchmark (point_bench.py) and its variants |
openpi_diagnostics/ |
Replay and determinism checks for the OpenPi pipeline |
workspace_outputs/experiments/... |
Small git-ignored outputs from the workspace (subgoal-label audit out/, planner GT labels, review videos, logs, eval report summaries) at their workspace paths |
hf download rchu233/kinder-ar-results --repo-type dataset --include "pi05_evals/kinder_subgoal_eef_b64_v1/*" --local-dir kinder-ar-results
- Downloads last month
- 194