Skip to content

Collect data

SimArena helps teams collect data from simulation runs and prepare it for robot learning.

  • Demonstration episodes from scenes.
  • Camera, depth, lidar, and IMU-style sensor outputs where configured.
  • Scenario outputs from scripts, including success and failure signals.

Built alongside LeRobot. LeRobot is the HuggingFace-maintained open-source toolkit and dataset format for robot learning — effectively the closest thing the open-source robotics community has to a shared standard for demonstrations and training pipelines. SimArena’s data pipeline is LeRobot-compatible by design, and CodecFlow co-founder Khalil collaborates with the LeRobot team directly. Episodes recorded in SimArena drop into LeRobot’s training pipelines without conversion, so data captured here lives in the same ecosystem the broader robotics research community already uses.

Good robot policies need useful data. SimArena keeps scene design, execution, data capture, and training setup close together so teams can iterate faster.