corl-experiments
Use when designing or auditing experiments for a CoRL robot-learning paper — seeds and evaluation-episode counts, task-suite breadth, real-robot versus simulation evidence, sim-to-real gap measurement, baseline fairness across BC/RL/VLA families, generalization splits, and statistics for success-rate claims.
Profile →npx skills add brycewang-stanford/Awesome-Journal-Skills --skill corl-experiments --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
# CoRL Experiments At CoRL the object under evaluation is a *learned policy*, which makes the evidence problem statistical twice over: training is stochastic (seeds, data order, initialization) and execution is stochastic (initial states, physics, sensor noise). An experimental design that controls only one of the two is the most common weakness this reviewer pool writes up. ## Match the evidence to the claim, not the venue | Claim in the paper | Minimum credible evidence shape | |---|---| | "Method X learns task family T" | Multiple training seeds; per-task success over many scripted-reset episodes | | "X outperforms baseline Y" | Same data, same evaluation protocol, same tuning effort for both; dispersion reported | | "X transfers sim-to-real" | The *same checkpoint* evaluated in sim and on hardware; the gap reported as a number | | "X generalizes to novel objects/scenes/instructions" | Held-out splits defined before training; per-split breakdown, not a pooled average | | "X scales with data" | ≥3 dataset sizes on the same axis; no two-point "trends" | | "X runs in real time on the robot" | Latency/frequency measured on the deployed compute, stated with hardware | The routing con
- Match the evidence to the claim, not the venue
- The two-layer randomness protocol
- Sim, real, and the space between
- Baseline fairness across method families
- Generalization splits that survive scrutiny
- Ablations and the Limitations coupling
- Design-review checklist
What does the corl-experiments skill do?
Use when designing or auditing experiments for a CoRL robot-learning paper — seeds and evaluation-episode counts, task-suite breadth, real-robot versus simulation evidence, sim-to-real gap measurement, baseline fairness across BC/RL/VLA families, generalization splits, and statistics for success-rate claims.
How do I install it?
Run `npx skills add brycewang-stanford/Awesome-Journal-Skills --skill corl-experiments --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From brycewang-stanford/Awesome-Journal-Skills, a repository with 909 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.