> For the complete documentation index, see [llms.txt](https://world-models-from-scratch.gitbook.io/wmfs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://world-models-from-scratch.gitbook.io/wmfs/chapters/chapter_01/02-why-we-need-world-models.md).

# 1.2 Why Do We Need World Models?

The value of a world model comes from looking beyond the present. It gives us a way to reason about how the world might change over time. An observation tells us what the world is like now, but choosing what to do may require knowing what could happen next.

Consider a robot at a junction. It can go left or right. Before the robot moves, the world model predicts a sequence of future positions for each path. Each predicted sequence is a rollout. The rollouts show how the two choices may develop over several steps.

## Using Rollouts to Choose an Action

To choose between the two paths, the rollouts must be compared. A planner performs this comparison. It assigns a score to each rollout.

For the robot, the rollout that reaches the goal in four moves receives a higher score than the rollout that requires seven moves.

The planner selects the path with the highest score and returns the first action needed to follow it. After receiving a new observation, the process repeats. This use of a world model is called model-based planning.

![](https://2690027296-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKukkmMWtxTgXSbpK92Uk%2Fuploads%2Fgit-blob-9f66e82a2cb8f3c8f7e5e210864ed256659152fa%2Ffig_1_4_reactive_vs_planning.png?alt=media)

*Figure 1.2: The planner proposes candidate plans. The world model predicts a rollout for each plan. The planner scores the rollouts and selects an action.*

The same process can be written as:

```python
def choose_action(model, observation, candidate_plans, score):
    results = []

    for plan in candidate_plans:
        future = model.rollout(observation, plan)
        results.append((score(future), plan))

    _, selected_plan = max(results, key=lambda item: item[0])
    return selected_plan[0]
```

## When One Plan Has Several Outcomes

The example above produces one rollout for each plan. A probabilistic world model can produce several rollouts for the same plan.

For example, a robot wheel may slip. An object may move by different amounts after a push. These differences produce a distribution of possible outcomes.

The planner can compare the average score across the sampled rollouts. It can also measure the probability of entering a specified state.

![](https://2690027296-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKukkmMWtxTgXSbpK92Uk%2Fuploads%2Fgit-blob-de991ec112505c8bccee3b36086b47bd7c2e715f%2Ffig_1_5_one_state_many_futures.png?alt=media)

*Figure 1.3: Each candidate plan produces a distribution of possible outcomes. The planner evaluates the distribution for each plan.*


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://world-models-from-scratch.gitbook.io/wmfs/chapters/chapter_01/02-why-we-need-world-models.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.