A research team led by Professor Lan Xuguang at Xi'an Jiaotong University has proposed AffordSim, a system that combines open-vocabulary 3D affordance prediction with simulated trajectory generation. Starting from natural language task descriptions, the system localizes functional regions on object surfaces and integrates motion planning to generate executable robotic manipulation demonstrations. The study builds a benchmark spanning 50 tasks, 7 categories of manipulation skills, and 5 types of robotic arms, together with an asset library of more than 500 rigid and articulated objects, and supports domain randomization operations such as real-scene background replacement and cross-camera viewpoint rendering. Experiments show that on medium- and high-difficulty tasks with stronger functional-region constraints, AffordSim achieves a markedly higher trajectory collection success rate than general grasping combined with semantic filtering; the collected simulated demonstrations can support zero-shot real-robot deployment of robotic policies, offering an effective path toward reducing manual tuning in data generation and evaluating how affordance understanding affects robotic manipulation in practice.