Start here · Understand the space · Updated 2026-08-23
For decades, robotics had a door policy: a lab, a budget, a degree. In the last two years the door fell off. This is the map of what is behind it and the ladder up.
Three things converged. Hardware collapsed in price: a capable leader and follower arm pair costs about $230 in parts, printable frame included. Models opened up: more than 80 robot brain models shipped in three years, and several of the good ones are free to download and fine-tune at home. And data standardized: one shared format means a dataset recorded in a bedroom in Lisbon trains a model in Seoul. Any one of these would matter; together they turned robotics into something a person can just start.
Four layers, top to bottom. Frontier labs building closed models on private data. Data companies paid to record for those labs. An open ecosystem, centered on the LeRobot community, where models, datasets, and tools are public. And the newcomers: thousands of people with cheap arms, recording data and training models at kitchen tables. The open layer is where you enter, and it is genuinely welcoming to beginners who show their work.
1: choose and build a cheap arm. 2: teleoperate until your hands are smooth. 3: record a small dataset with real labels. 4: lint it, license it, share it. 5: fine-tune an open model on your episodes. 6: watch your robot do the task alone, then iterate. Every rung has a plain-English guide in this series, starting at choosing your arm. People climb the whole ladder in a few weekends.
Rung 6 is a beginning. From there people branch: some become serious dataset contributors whose data trains open models. Some run evaluation rigs when real-robot testing networks open (that is coming: The Proving Ground). Some turn the skills into jobs; today's kitchen-table fine-tuners are next year's robot company ML engineers. And some just enjoy owning the strangest hobby on their street.
Robots are slower and clumsier in person than in launch videos; simulation demos overpromise; and a third of the community's popular datasets fail basic quality checks, which is why we built the finder. Enter anyway. The gap between what exists and what is promised is exactly where newcomers get to matter.