What clusters does this skill support?

The skill is pre-configured for the logical 'lrz cluster' flow, which includes dual-submission to lrz_x100 and mcml_x100 partitions with automated conflict resolution.

How are post-training rollouts handled?

Rollouts are treated as evaluation events. Instead of creating new runs, results are appended to the original Stage 3 training note to maintain a clean experiment lineage.

How does it handle documentation for major runs?

It creates a dated experiment config in 'config/experiment/runs/' and a matching markdown note in 'docs/runs/', recording git commits, config deltas, and artifact paths.

Can I use this for quick debugging?

Yes. By explicitly stating 'test' or 'debug', the skill skips the formal documentation flow and uses CLI overrides for faster iteration.

What environment is used for local runs?

The skill automatically selects the appropriate Conda environment based on the stage: 'hlrp' for Stage 1 and 2, and 'lerobot' for Stage 3 and evaluation rollouts.

Robotic Learning Run Manager

Name: Robotic Learning Run Manager
Author: felixmin

byfelixmin

0•

데이터 과학 및 ML

Orchestrates multi-stage robot learning experiments across local workstations and high-performance compute clusters.

This skill automates the complex lifecycle of robot learning runs, specifically optimized for the Hierarchical Learning for Robotic Policies (HLRP) framework. It streamlines the transition between local development and cluster-scale training (Stages 1-3) while enforcing rigorous documentation standards. By managing job submissions to clusters like LRZ and MCML, implementing a dual-submit 'race' strategy to minimize queue times, and maintaining precise artifact provenance, it ensures research experiments are both reproducible and efficient.

주요 기능

01Dual-cluster job submission with intelligent 'predicted-start' monitoring.

02Automatic generation of dated experiment configs and markdown documentation.

03Automated multi-stage training workflows (LAM, Policy, and LeRobot).

04Integrated checkpoint provenance and upstream artifact tracking.

05Persistent local execution management using tmux sessions.

060 GitHub stars

사용 사례

01Launching major Stage 3 training runs on remote GPU clusters with full documentation.

02Attaching post-training rollout evaluations and metrics to existing experiment notes.

03Running lightweight local debug tests using direct CLI overrides without config bloat.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add felixmin/robot-learning-from-video start-run

For use in Claude.ai and ChatGPT

주요 기능

01Dual-cluster job submission with intelligent 'predicted-start' monitoring.

02Automatic generation of dated experiment configs and markdown documentation.

03Automated multi-stage training workflows (LAM, Policy, and LeRobot).

04Integrated checkpoint provenance and upstream artifact tracking.

05Persistent local execution management using tmux sessions.

060 GitHub stars

사용 사례

01Launching major Stage 3 training runs on remote GPU clusters with full documentation.

02Attaching post-training rollout evaluations and metrics to existing experiment notes.

03Running lightweight local debug tests using direct CLI overrides without config bloat.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add felixmin/robot-learning-from-video start-run

For use in Claude.ai and ChatGPT