Enhancing Game-Server AI with Distributed Client Computation

In the context of online role-playing games, we evaluate offloading AI computation from game servers to game clients. In this way, the aggregate resources of thousands of participating client machines can enhance game realism in a way that would be prohibitively expensive on a central server. Because offloading can add significant latency to a computation normally executing within a game server’s main loop, we introduce the mechanism of AI partitioning: splitting an AI into a high-frequency but computationally simple component on the server, and a lowfrequency but computationally intensive component offloaded to a client. By designing the client-side component to be stateless and deterministic, this approach also facilitates rapid handoff, preemptive migration, and replication, which can address the problems of client failure and exploitation. To explore this approach, we develop an improved AI for tactical navigation, a challenging task to offload because it is highly sensitive to latency. Our improvement is based on calculating influence fields, partitioned into server-side and client-side components by means of a Taylor series approximation. Experiments on a Quake-based prototype demonstrate that this approach can substantially improve the AI’s abilities, even with server-clientserver latencies up to one second.