The paper introduces CooHOI (Cooperative Human-Object Interaction), a novel framework designed to address multi-character collaboration in transporting objects. The framework employs a two-phase learning paradigm: initial skill acquisition for single agents using the Adversarial Motion Priors (AMP) framework, followed by transfer learning to enable collaborative strategies using Multi-Agent Proximal Policy Optimization (MAPPO). This approach leverages the dynamics of the manipulated object as feedback, facilitating implicit communication and coordination among agents. The framework is efficient, does not rely on extensive multi-character interaction data, and can be extended to handle different object types and numbers of participants. Experiments demonstrate that CooHOI enables humanoid robots to perform natural and precise object-carrying tasks in both single-agent and multi-agent settings, with improved success rates and transportation distances compared to baseline methods. The framework's effectiveness is validated through comprehensive evaluations on various object categories and detailed ablation studies, highlighting the importance of different design choices. Future work will focus on integrating dexterous hands and active perception to enhance the framework's capabilities.The paper introduces CooHOI (Cooperative Human-Object Interaction), a novel framework designed to address multi-character collaboration in transporting objects. The framework employs a two-phase learning paradigm: initial skill acquisition for single agents using the Adversarial Motion Priors (AMP) framework, followed by transfer learning to enable collaborative strategies using Multi-Agent Proximal Policy Optimization (MAPPO). This approach leverages the dynamics of the manipulated object as feedback, facilitating implicit communication and coordination among agents. The framework is efficient, does not rely on extensive multi-character interaction data, and can be extended to handle different object types and numbers of participants. Experiments demonstrate that CooHOI enables humanoid robots to perform natural and precise object-carrying tasks in both single-agent and multi-agent settings, with improved success rates and transportation distances compared to baseline methods. The framework's effectiveness is validated through comprehensive evaluations on various object categories and detailed ablation studies, highlighting the importance of different design choices. Future work will focus on integrating dexterous hands and active perception to enhance the framework's capabilities.