skills.rest
junhyekh
junhyekhCOMMUNITY
Explore all skills by this author →
GitHub

rlxp-dr-design

Design domain randomization ranges and schedules for reinforcement learning experiments.

1|Updated May 14, 2026
Search Tags:#robustness#reinforcement learning#domain randomization#sim-to-real#probe analysis#experiment tuning
One-click install
npx skills add https://github.com/junhyekh/rlxp --skill rlxp-dr-design
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rlxp-dr-design
Source: https://github.com/junhyekh/rlxp/tree/main/plugins/rl-experiment-assistant/skills/rlxp-dr-design
Command: npx skills add https://github.com/junhyekh/rlxp --skill rlxp-dr-design

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you decide how to tune domain randomization when an RL policy performs well in nominal conditions but becomes brittle under randomized, held-out, or sim-to-real evaluation.

Core Features & Use Cases

  • Robustness diagnosis: Compares nominal and randomized results to identify whether dynamics, observation, terrain, asset, latency, or actuation effects are causing the performance gap.
  • Probe-backed tuning: Recommends controlled DR probes and uses their measured bounds to expand, contract, schedule, or rebalance randomization safely.
  • Study-safe recommendations: Keeps evaluation distributions stable unless a new study is created, and avoids inventing broad ranges without evidence.
  • Use case: A robot policy handles the training environment but fails on unseen surface friction, so the Skill proposes a targeted probe plan and a constrained DR update.

Quick Start

Use the rlxp-dr-design skill to review the current DR setup, compare robustness evidence, and propose a probe-backed tuning plan.

Frequently Asked Questions about rlxp-dr-design

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I tune domain randomization when my reinforcement learning policy is brittle during sim-to-real transfer?▼

Tune domain randomization by comparing nominal and randomized evaluation results to diagnose brittleness, then applying probe-backed adjustments to expand or contract randomization ranges safely. This ensures your reinforcement learning policy generalizes better during sim-to-real transfer.

What causes performance gaps between nominal training and held-out generalization in RL?▼

Performance gaps in held-out generalization are caused by mismatches in dynamics, observation, terrain, asset, latency, or actuation effects between nominal and randomized environments. Comparing these specific results helps identify which randomization factor is degrading policy robustness.

How do I safely expand domain randomization ranges without breaking existing RL performance?▼

Safely expand domain randomization ranges by running controlled DR probes to measure bounds, then using that evidence to adjust ranges or schedules. This prevents inventing broad ranges without evidence and keeps evaluation distributions stable unless a new study is created.

What data do I need to design a domain randomization tuning plan for a robotics policy?▼

To design a domain randomization tuning plan, you need DR configs, nominal and randomized results, train and held-out comparisons, probe summaries, task cards, and contract scope. These inputs define safe probe-backed adjustments for your robotics policy.

When should I not use broad domain randomization ranges for sim-to-real reinforcement learning?▼

You should avoid broad domain randomization ranges when you lack measured evidence from controlled probes. Expanding ranges without probe data can destabilize sim-to-real reinforcement learning, so recommendations must be study-safe and backed by measured bounds.

Can I schedule domain randomization adjustments instead of applying them all at once?▼

Yes, you can schedule domain randomization adjustments. The tuning process uses probe analysis to recommend whether to expand, contract, schedule, or rebalance randomization ranges, allowing gradual adjustments based on measured evidence rather than immediate broad changes.

Related Skills

Explore More Software Engineering Skills→
junhyekhjunhyekh
1

rlxp-dr-probe

Plan one-parameter-at-a-time domain-randomization probes for incumbent reinforcement-learning policies.

Community
Intermediate
robotics-playgroundrobotics-playground

isaac-rl

Guides RL locomotion policy development for Unitree G1 using NVIDIA Isaac Sim and Isaac Lab.

Official
Advanced
ragnarokhaaragnarokhaa
4

hermes-atropos-environments

Construct and test Hermes Agent reinforcement learning environments for Atropos training.

Community
Advanced
RebelHawk-TKRebelHawk-TK

positioning-workshop

Guide product teams through a structured positioning discovery workflow.

Community
Intermediate