Multi-agent consensus using both current and outdated states