Rollout
Also called Trajectory
In reinforcement learning, a rollout is a sampled sequence of interaction: observations or states, actions, and resulting rewards. It provides experience for analysis or learning.
[OpenAI]In practice · hypothetical example
A simulated robot’s trial records each position, action, and reward until it reaches a goal.
[OpenAI]A little deeper
An episode can end at a terminal condition, while a collected rollout may also stop at a chosen horizon. Here the word refers to sampled behavior, not deploying software. [OpenAI]
A common mix-up
A rollout is the same as releasing a new model to users.
The term has a separate deployment meaning; this entry covers interaction trajectories. [OpenAI]
What is recorded in an RL rollout?
Sources & editorial notes
Evidence: supported. Primary-source support for this scoped entry; publication approved by the project owner.
- Part 1: Key Concepts in RL ↗ (opens in new tab)OpenAI · Publication date unknown
Relevant section: Trajectories; Reward and Return
Last editorial review: 2026-09-13 by project-owner.
First observed in this corpus: Unknown.
Revision history
Revision 2 · Created 2026-09-13 · Updated 2026-09-13
Project owner approved the current content for publication. Existing evidence scope and limitations remain applicable.