CameraBots: Cinematography for Games with Non-Player Characters as Camera Operators

This paper proposes CameraBots, autonomous virtual camera operators modelled on the non-player character bots already common in 3D games, as a way to bring cinematographic principles into real-time game camera work. The authors argue that conventional game cameras, typically a single first-person or over-the-shoulder view, ignore the shot-selection, cutting, and framing conventions that film practice uses to make footage engaging and coherent. Their central thesis is that because virtual cameras are not bound by physical limits, and because bot code for navigation, spatial reasoning, and path following already exists, multiple cinematographic camera operators can be built with relatively little new work.

The proposed architecture mirrors a film set. A director module evaluates the action several times per second and selects characters and props to film, while a cinematographer module assigns specialised CameraBots to produce particular shot types: establishing, character (long and medium), close-up, first-person, and over-the-shoulder. The paper includes pseudocode describing how the cinematographer chooses among these based on whether the player's avatar or another principal character is attacking, moving, or stationary, and it stresses that the modular design allows new CameraBot types to be added and the system to be ported to other engines through an API. The authors position the approach not only for single-player games, where it could briefly take camera control away from the player to advance the story, but also for spectator views, replays, and machinima.

The work is conceptual and at an early stage. The authors report that they are implementing the system as a modification to the Quake II engine in C++, and they plan future user testing for playability and visual immersion, but the paper presents no experimental results, user studies, or quantitative evaluation of the claimed benefits in engagement or reduced player disorientation. The literature review draws on prior work in automatic camera control for interactive narratives, learning environments, and games, and situates the contribution as the use of multiple cameras with partial control of the action, which the authors identify as a gap in existing research.