Gamer: A Generalist Vision-Language-Action Foundation Model for Long-Horizon Real-Time Game Agent
The authors report skilled-player-level operational efficiency in several 3D open-world games after training on tens of thousands of hours of keyboard and mouse interaction.
Gamer combines perception, memory management and multi-step reasoning in one architecture for 2D interfaces and 3D worlds. Its “sparse-thinking” training reduces redundant computation while preserving reasoning for complex decisions. Multi-turn reinforcement learning addresses finite context windows, accumulated errors and sparse rewards during long tasks. A cloud-edge inference engine supports concurrent game environments and low-latency interaction; the abstract gives no numerical results.
ArXiv · AI/CL/LG's note
Gamer combines perception, memory management and multi-step reasoning in one architecture for 2D interfaces and 3D worlds. Its “sparse-thinking” training reduces redundant computation while preserving reasoning for complex decisions. Multi-turn reinforcement learning addresses finite context windows, accumulated errors and sparse rewards during long tasks. A cloud-edge inference engine supports concurrent game environments and low-latency interaction; the abstract gives no numerical results.
ArXiv · AI/CL/LG's note
score 5