Intent-Based Human-Object Interaction Driven Autonomous Task Acquisition Method for Robots

Embodied intelligence achieves perception, decisionmaking, and execution through interaction with the environment via physical embodiments. However, current systems often rely on manually predefined tasks, which can lead to task interruption or uncontrollable consequences when the environment does not meet the conditions for task execution. To address this issue, this study proposes an intent-based human-object interaction-driven autonomous task acquisition method for robots, enabling robots to dynamically generate reasonable and executable tasks based on human intentions and the state of the environment. First, a method is designed to recognize human interaction intentions by predicting hand motion trajectories. The future motion trajectory of the human hand is predicted by tracking changes in hand key points, and by considering the spatial positions of objects in the scene, potential interactions between humans and objects are identified, thus clarifying human intentions. Next, a task acquisition method is proposed based on intention and object geometric features. Using the identified human intentions, task goals are obtained, and combined with the three-dimensional geometric features of objects, commonsense reasoning is used to automatically determine the task type, allowing for the execution of appropriate tasks for the actual situation. This method enables the robot to autonomously perceive and acquire human intentions in dynamic environments, achieving intelligent and autonomous task acquisition with high flexibility and success rates.