Training

Training sits at the end of the SimArena loop: build a scene, run it, collect data, then use that data to improve robot behavior.

Training inputs

• Demonstration episodes.
• Sensor outputs from simulation.
• Scenario metadata and success checks.

Training workflows

CodecFlow is built for imitation learning and reinforcement learning workflows. Teams can use collected data to prepare training runs, evaluate policies, and repeat the loop with updated scenes or tasks.

The useful loop

Build the task, collect data, train, test again. SimArena keeps that loop in one workspace.