In this letter, we propose a motion generation strategy using imitation learning (IL) for biomimetic robots to interact with animals in a more natural manner. We previously developed a bioinspired spine mechanism allowing the robot to mimic rat motion closely. Thanks to the spine morphology, we established the corresponding relationship of motion vector (spatial location and positions of key movement joints) between the robot and rats. We denote the motion vectors of two agents (demo robot and policy robot) as the interaction motion features (IMFs), where the motion vector of demo robot is directly copied from the demo rat and that of policy robot is generated using IL. First, we trained multilayer perceptrons by inputting the IMFs from the rat-rat interaction to generate the predicted IMFs for robot-robot interaction. Secondly, robot-robot IMFs are used to train a policy net, whose cost is the mean squared errors between the predicated and current features. Finally, we evaluated the proposed method in a simulation environment for the interaction between demo robot and policy robot. The results show effective interaction duration between robots has an increment of 16% than that between rats. Moreover, it also has a better concentration span than that of rat-rat interaction.
Current artificial intelligence evaluation often relies on static benchmarks and task-specific performance metrics, which are useful for system comparison but limited in assessing developmental progression, cognitive architecture, adaptive transfer, and real-world applicability. Building on developmental approaches to AGI testing, including our earlier conceptual proposal, this article develops an operational formulation of the General–Specialized–Applicable (GSA) framework for artificial general intelligence (AGI) evaluation. The framework organizes evidence into three non-interchangeable stages: the General stage, which assesses foundational capacities for open-ended generalization, value-oriented regulation, and autonomy; the Specialized stage, which evaluates stable domain-specific competence; and the Applicable stage, which examines whether such competence can be deployed safely and robustly in realistic environments. The article further introduces a dynamic task-generation pipeline and a provisional operational rubric for interpreting stage-specific evidence. An illustrative case study using household tasks in a simulated embodied environment illustrates how GSA can provide diagnostic information beyond aggregate benchmark scores by identifying capability gaps in current multimodal large language model (MLLM) agents. Rather than offering a final universal standard, the GSA framework provides an evaluation-oriented structure for connecting benchmark performance with architectural readiness, specialization, and deployment-level applicability.