Crowdsourcing Controller -- Utilizing Reliable Agents in a Multiplayer Game

TL;DR

This paper introduces a novel Crowdsourcing Controller for multiplayer games that dynamically groups player inputs and weights them by estimated reliability, improving game performance and accurately assessing player skills.

Contribution

It proposes a dynamic input grouping method and a reliability-based weighted voting system, addressing limitations of static frames and uniform weighting in previous approaches.

Findings

Higher game scores with the Crowdsourcing Controller

Effective estimation and updating of player reliability scores

Improved multiplayer control through dynamic input and reliability weighting

Abstract

This paper presents a new use case for continuous crowdsourcing, where multiple players collectively control a single character in a video game. Similar approaches have already been proposed, but they suffer from certain limitations: (1) they simply consider static time frames to group real-time inputs from multiple players; (2) then they aggregate inputs with simple majority vote, i.e., each player is uniformly weighted. We present a continuous crowdsourcing multiplayer game equipped with our Crowdsourcing Controller. The Crowdsourcing Controller addresses the above-mentioned limitations: (1) our Dynamic Input Frame approach groups incoming players' input in real-time by dynamically adjusting the frame length; (2) our Continuous Reliability System estimates players' skills by assigning them a reliability score, which is later used in a weighted majority vote to aggregate the final output command. We evaluated our Crowdsourcing Controller offline with simulated players and online with real players. Offline and online experiments show that both components of our Crowdsourcing Controller lead to higher game scores, i.e., longer playing time. Moreover, the Crowdsourcing Controller is able to correctly estimate and update players' reliability scores.