About & methodology
Why this site exists
Robot learning moves fast and most of what gets written about it is either a paper nobody outside the field can read, or a demo video. If you are an engineer deciding whether to put a learned policy on a production line, or a researcher choosing a platform, neither helps much.
RobotLearningHub keeps one table: which algorithm was tested on which robot for which task, what the result was, and how far from production it was. Every row links to its source. Every row is reviewed by a person before it is published.
Who runs it
[OWNER TO WRITE — short bio and what you work on day to day.]
Maturity levels
Every result is tagged with the setting it was shown in. Higher is not “better research”; it is closer to something you can rely on.
- Simulation
- The result exists only in a simulator. Useful for comparing methods, not proof it works on hardware.
- Lab
- Run on a real robot, in a research lab, on a tabletop or a staged scene. Most papers live here.
- Pilot
- Run on a real robot in a real workplace for a limited time or scope, usually with engineers nearby.
- Production
- Running in a real workplace, unattended, for weeks or months, with someone depending on it.
Evidence grades
Maturity says where a result happened. Grade says how much you should trust the number.
- Astrong
- Real robot, at least 50 rollouts or a multi-day deployment, and a clear metric.
- Bmoderate
- Real robot, but a small number of trials or an unclear metric.
- Cweak
- Simulation only, or anecdotal / press coverage without numbers.
Numbers are copied from the source as reported. When a source does not give a number, the field is left empty rather than estimated. When trial counts are small or metrics unclear, the grade says so.
Independence
No robot or software vendor pays for placement, and there are no affiliate links. The site is funded by consulting: when a team needs a decision for their own line or lab, they can hire the person who maintains this table.
Corrections
If a row is wrong or a source is misread, say so. Corrections are made quickly and the row keeps its history.