BRISA Reading Group 01 - Play2Perfect

​Before the official launch this fall, we're opening with the thing we want the whole community built around: sitting down with a paper, understanding it properly and asking what it would take to build it ourselves.

​Our first paper is Play2Perfect: What Matters in Dexterous Play Pretraining for Precise Assembly? (Lum, Kedia, Liu & Bohg, 2026) out of Stanford and Cornell, with videos and open code on the project page.

​Precise assembly with a five-fingered hand is difficult for two reasons: the tasks are contact-rich, so teleoperating demonstrations is inherently challenging. Additionally, they are sparse-reward, so RL exploration never gets a grasp of success.

​An answer might come from the engineering perspective, for instance through building custom grippers, fixtures and tool attachments. However, the authors take a different path: before a robot can perfect assembly, it should first learn to play.

​First, a single policy learns in simulation to pick up a thousand random objects and move them to random target poses. No demonstrations and cues on what the real task will be.

​Then that same policy is finetuned on the actual assembly, told nothing except whether it succeeded.

​It then runs on a real robot arm and hand with no further training, fitting a T-shaped block into a hole half a millimetre wider than the block itself.

​We selected this paper for our first session for a very specific reason. Everything needed is there: object generation, reward terms, hyperparameters, the full compute setup, 3D printable parts, code on GitHub and videos.

​We have seen the potential of this paper - with all its auxiliary material, a team could take it from the first line to a working system.

​This gap - between reading a robotics paper and building the thing - is what BRISA exists to bridge.

​No preparation needed. Skim the abstract if you have ten minutes, watch the videos if you have two. Everyone is welcome - whatever you study, wherever you study and however much robotics you have done so far.

​Come with questions, with disagreements, or with something you want to build.

​🎟️ There are only a limited number of spots, so we can keep the session personal and interactive.

​📍 Hosted at the Merantix AI Campus. Doors open 18:30, we start shortly after.

​👋 New here? Join the community group — everything else gets announced there first.