Is there a bottleneck which would hinder putting this architecture in charge of a humanoid? Would it be able to operate continuously, for example in conjunction with an LLM for long-term reasoning? Doom seemingly works extremely well.