Researchers propose PACT (Phenotype-Aware Contrastive Team Representation) to solve a challenge in multi-agent systems where controlled agents must collaborate with unfamiliar teammates of diverse coordination phenotypes. PACT uses phenotype-aware contrastive learning and relational reasoning to distinguish between these phenotypes and capture inter-agent interactions. The method is tested on various collaboration tasks, showing significant improvements over existing methods in terms of out-of-distribution evaluation and sample efficiency. --- Why it matters: This matters because it addresses a key challenge in multi-agent systems where controlled agents must adapt to new teammates with different coordination styles. Solving this problem can improve the performance of autonomous teams in real-world applications such as robotics, logistics, or human-robot collaboration. Source: https://arxiv.org/abs/2510.25340