My kid has been rewatching Avatar: The Last Airbender, which means I have been rewatching it, and that gave me a roleplaying game related idea, not so much in regard to its content, but really its structure.
The creation of the fantasy epic, such a staple of fantasy media, has always been hard in games because historically, attempts to do so have led to drastically limited options for player agency. At best, the Adventure Path that is the modern descendant of the Dragonlance modules tends to be really linear. At worst, it's an outright railroad.
I don't think it has to be that way, though, but it would require some discussion and buy-in from players and a good session zero. Here's how I think it could work:
1. The GM tells the players the campaign setting and situation and suggests (but not mandates) a Quest, perhaps. Or perhaps, the players and the GM sort of make that up together? The "Quest" is the desired outcome: defeat the Firelord in the case of The Last Airbender or defeat Sauron in Lord of the Rings.
2. The player's make up characters, finalize the Quest, and plan the steps they think they will need to achieve it. The Quest needn't be etched in stone. It's possible the campaign as it unfolds might lead to a different goal, e.g.: Babylon 5 was our last, best hope for peace. It failed. But in the year of the Shadow War, it became something greater: our last, best hope for victory. It's even conceivable PCs might switch sides. Anyway, there should also be more character specific goals woven in, not just big campaign ones.
3. The GM plots those steps both geographically on a pointcrawl map and node-wise for a campaign structure map and makes clocks of antagonist/rival actions and other events. It's important to note here that the steps which will become nodes aren't plotted scenes. They aren't linked to each other in a linearly (or strictly linear) fashion for the most part, and they aren't supposed to go any certain way. Nothing is "supposed" to happen. In Avatar, Aang has to master the 4 elements. That goal could have played out in a lot of different ways. In fact, it takes two potential teachers before he ultimately gets to learn firebending. Localizing potential places where the goals can be achieved is important, because fantasy epics tends to cover a lot of geography. They aren't just dramas or soap operas to be played out in a limited location.
4. The players choose where to go and have other adventures and encounters along the way due to those choices. This may call for a bit of separation of player and character knowledge, but even without that, I feel like it works if the players just know the likely location of achieving one of their goals. Circumstances may mean it doesn't work out. The world doesn't stay static. But any unsuccessful attempt to achieve a goal at a point should always yield clues to a goal--either another one or the one they failed to achieve. In this sense, it's like running a mystery; clues to the next goal location shouldn't be hard to find.
5. Players can alter goals in response to events or their desires. New point crawl "maps" may need to be generated in response. When new goal nodes come online, new hooks and areas of interest need to be populated around them. It's the "story" goals embedded in sandboxy locations that makes this much less linear than an adventure path.
6. Repeat until the PCs achieve the goal or the clocks expire and a new status quo (and possibly campaign) is established. What if the hobbits fail to destroy the ring before Sauron's victory? Well, the story needn't be over.
This approach doesn't feature the degree of session to session freedom of the completely sandbox game, it's true. However, the player collaboration in the planning phase ensures it's not a GM enforced story. Indeed, both players and GM will be surprised by the final shape of the emergent story.
While this may be a bit of a novel approach (at least I haven't seen anyone ever talk about it) ideas about "node-based scenario design" and "mission-based adventures" have existed for a long time. What this does to enhance those is get player input prior to the missions and link the nodes in a grander campaign.