Human-like Bots for Tactical Shooters Using Compute-Efficient Sensors

TL;DR

The paper tackles the challenge of deploying human-like AI bots in commercial tactical shooters under CPU-bound constraints by proposing compute-efficient, pixel-free imitation learning using a minimal ray-cast sensor suite. It introduces Lyra:Ascent, a VALORANT-inspired testbed, and trains memory-enabled, light-weight networks via Behavior Cloning on rich human-trajectory datasets, achieving CPU-time per decision in the single-digit millisecond range. Model D emerges as the best overall balance of realism and efficiency, closely matching human behavior across distributions and spatial patterns while remaining practical for industry deployment; human evaluation indicates substantial believability with some remaining gaps in movement realism. The work demonstrates a viable path from academic methods to production-ready, human-like bots and outlines clear avenues (RL, GAIL, transformer-based offline methods) for further improving robustness and diversity of agent behavior in multiplayer games.

Abstract

Artificial intelligence (AI) has enabled agents to master complex video games, from first-person shooters like Counter-Strike to real-time strategy games such as StarCraft II and racing games like Gran Turismo. While these achievements are notable, applying these AI methods in commercial video game production remains challenging due to computational constraints. In commercial scenarios, the majority of computational resources are allocated to 3D rendering, leaving limited capacity for AI methods, which often demand high computational power, particularly those relying on pixel-based sensors. Moreover, the gaming industry prioritizes creating human-like behavior in AI agents to enhance player experience, unlike academic models that focus on maximizing game performance. This paper introduces a novel methodology for training neural networks via imitation learning to play a complex, commercial-standard, VALORANT-like 2v2 tactical shooter game, requiring only modest CPU hardware during inference. Our approach leverages an innovative, pixel-free perception architecture using a small set of ray-cast sensors, which capture essential spatial information efficiently. These sensors allow AI to perform competently without the computational overhead of traditional methods. Models are trained to mimic human behavior using supervised learning on human trajectory data, resulting in realistic and engaging AI agents. Human evaluation tests confirm that our AI agents provide human-like gameplay experiences while operating efficiently under computational constraints. This offers a significant advancement in AI model development for tactical shooter games and possibly other genres.

Figures (15)

• Figure 1: Selected screenshots of the Lyra:Ascent game with (a) defending player at the start point, (b) a player planting the bomb, and (c) an area affected by a grenade.
• Figure 2: The Lyra:Ascent map used for our experiments. The map is heavily inspired by the Ascent level of VALORANT (Riot Games, 2020). (a) Shows a view of the level created in Lyra:Ascent, and (b) shows a map schematic of the level where the A represents the attacker spawn point, the D represents the defender spawn point, and the B is the bomb site.
• Figure 3: The figure illustrates (a) 15 horizontal and (b) 15 vertical rays that are distributed non-uniformly, forming a $15 \times 15$ grid-based sensory input.
• Figure 4: AKDR vs average damage dealt per round by the player.
• Figure 5: Pitch and Yaw distributions of the players.