Analyzes, summarizes, and documents robot learning experiment runs across local and cluster environments.
The evaluate-run skill automates the inspection of Hierarchical Learning for Robot Policy (HLRP) experiments, providing instant summaries of training losses, success metrics, and rollout performance. It intelligently resolves run references from Slurm job IDs, directory paths, or configuration files, and automatically synchronizes findings to the project's documentation. Whether monitoring an active cluster job via SSH or finalizing results for research, this skill streamlines the experiment analysis workflow for robot learning researchers by extracting insights from logs, checkpoints, and Hydra configurations.
Características Principales
01Multi-source run resolution including Slurm IDs, local paths, and 'check last run' logic
02Cross-platform inspection with automatic SSH cluster access for remote jobs
03Stage-specific evaluation for training (Stage 1/2) and rollout (Stage 3) metrics
04Automated documentation updates for finalized experimental runs in markdown format
05Real-time status tracking and metric extraction from logs, Hydra, and LeRobot outputs
060 GitHub stars
Casos de Uso
01Monitoring live training progress and loss trends on a Slurm-managed compute cluster
02Quickly identifying the causes of failed runs by inspecting .out and .err logs alongside config snapshots
03Generating automated summary tables for robot rollout success rates in experiment documentation
