Abstract: Obstacle avoidance is an extremely challenging issue faced by crowd simulation. Many different approaches have been proposed with the aim of avoiding collision and eliminating intersection artifacts between two agents or between agent and environment. But most of these methods can not guarantee against overlapping, while other methods which can ensure no overlap have many space restrictions on agent's behavior or may cause visually unpleasant artifacts. To solve this problem, hierarchical obstacle avoidance focuses on a three levels obstacle avoidance method which contains two-level obstacle avoidance behavior and an intersection elimination level. The two-level obstacle avoidance behavior is used for avoiding collision during the behavior planning and execution while the intersection elimination level is employed to adjust overlapping after coarse renewed positions are calculated for agents. The two-level obstacle avoidance behavior, including static obstacle avoidance and dynamic obstacle avoidance, extremely reduces the complexity of the situation which should be considered in the planning and execution of agent's behavior by dividing the objects into two kinds: static obstacles and dynamic obstacles according to their attributions and making use of separate obstacle avoidance method to avoid collision. The intersection elimination level, based on agent's variable bounding box and its previous position, eliminates intersection artifacts absolutely in every update period without space restrictions and visual artifacts.