Grasping operations that fulfill object functional requirements endow robotic manipulation with practical significance. This work proposes a novel strategy framework for guiding dexterous hands to approach target states collision-free, via contact point-driven manipulation and hand-object interaction characterization. First, we semantically parameterizes finger links of the dexterous hand, to generate human-like grasp configurations capable of grasping and manipulating objects, based on functional requirement-oriented contact point set. Second, an Interaction Bisector Surface based hand-object spatial representation is adopted, to facilitate inverse motion planning from known contact states to non-contact configurations. Through reinforcement learning, we achieve inverse grasp trajectory planning to guide the generation of forward grasping motions. Finally, a high-DoF arm-dexterous hand simulation system is implemented, validating the method’s efficacy across diverse objects. Experimental results demonstrate that our strategy effectively generates grasp configurations, conforming to common functional intuitions while producing high-quality dexterous trajectories.