An arXiv preprint describes a planning approach for embodied agents that must cooperate without being told exactly what cooperation requires. The abstract notes that embodied agents are increasingly capable of perceiving, planning, and acting autonomously in dynamic environments, and perform well on complex cooperative tasks. It also says that recent advances typically assume that cooperative… — the abstract cuts off there, so the full assumption is not stated in the source.
The paper's title identifies the proposed solution as Discovery-Probability Graph Planning, aimed at tasks with hidden cooperation requirements. Based on the title and the opening sentences, the method appears to use discovery probabilities to guide planning when the cooperative structure of a task is not explicitly given. Because the source is truncated, the specific algorithm and any experimental results are not available.
The significance is practical: real cooperative tasks often involve unstated or hidden requirements, and agents that assume full knowledge may fail. This work points toward planning under that uncertainty, though the details remain to be read in the full paper.