Mosaic Cuts Multi-Agent Embodied Planning Latency by Attacking Failed Actions, Not Just Coordination

arXiv·medium signal

Mosaic identifies failed actions — stemming from inaccurate state tracking under partial observability and inefficient coordination — as the dominant latency bottleneck in LLM-based multi-agent embodied planning, rather than raw model speed. It targets those failures to make runtime-efficient planning practical, addressing a core reason such systems have stayed too slow to deploy. Relevant framing for anyone building embodied or long-horizon agent coordination.