Direct avatar control

The player maps controller or keyboard input to one character's locomotion (walk, run, jump, swim, fly) and interactions (pick up, use, attack, build) in real time. State owned: the avatar's position, velocity, facing, stance, and interaction target. Update rule: each frame, read input, apply a movement controller (kinematic or physics-based) subject to collision and terrain, and resolve interaction requests against objects in range. The world simulation runs around the avatar, often with only the area near it simulated in detail. It contrasts with overseer control, where the player commands at a distance. Canonical forms: Minecraft, Stardew Valley, The Long Dark, Subnautica, Valheim, and the farmer in Harvest Moon; hybrid games such as Satisfactory and Factorio put a factory builder in an avatar.

The player is in the world rather than above it. Scale, distance, and danger are felt in the body: a long walk to a resource, a fall, darkness. Systems are learned by touching them, and attachment shifts from managed agents to one's own character.

• movement speed, traversal options, and fast travel, which set how large the world feels
• interaction reach and whether actions take time or animate
• physics-based versus kinematic controller
• whether the avatar has needs or survival meters
• camera: first person, third person, or top-down
• how much of the world simulates outside the avatar's vicinity

• Large worlds with slow movement turn logistics into walking chores until fast travel or automation unlocks.
• Simulation limited to the avatar's surroundings freezes distant systems, breaking the illusion that the world runs without the player.
• Avatar control competes with overseer tools in hybrid games, forcing awkward building from a character's viewpoint.

• The avatar's travel time becomes the hidden limiting resource, pushing players to build transport and automation.
• Embodied presence makes environmental systems such as weather, temperature, and day-night directly threatening.
• Avatar location sets what is simulated, so player movement shapes world state in games that simulate only loaded chunks.

• Can learned motion models (physics-based character controllers trained by RL) give avatars natural interaction with arbitrary simulated objects without per-object animation?
• In generative world models where the world is rendered from the avatar's action stream, what simulation persistence can survive outside the view?