Abstract

Shared control is an assistive technology developed to increase access to console and PC video games. It allows to combine the input of two controllers so that a user (the pilot) can play a game while being supported by another user (the copilot). For example, in shooter game, the pilot might not be able to control all actions and can hence delegate the copiot to aim and to change the weapon. In our research, we are investigating how shared control systems are used by people with disabilities. One major limitation that emerged from our study is that the copilot might not be always available to play when the pilot wants to play. To address this problem, we are investigating how to automate the copilot through automatic agents.

System prototype

Several scientific papers evaluated partial automation approaches by designing ad-hoc games. Instead, we aim to enable partial automation on existing games. So, we designed a middleware layer that interfaces between the operating system and the game, combining the player's input with the agent's input. Thanks to this modular architecture, we can design autonomous agents and experiment them with existing games. We have designed two agents, a rule-based one and an AI-based one.

Rule-based agent

The rule-based agent can support players in the Doom game by providing support to:

• hold the run button
• automatic shoot
• automatic aim

AI-based agent

We are training an AI agent to control some of the commands of the Rocket League game. Currently the agent is able to:

• Accelerate and decelerate
• Use handbrake and boost
• Steer