In traditional crowd simulation, global path planning (GPP) and local collision avoidance (LCA) have been used to advance pedestrians toward their own goals without colliding. However, we found that using those methods in bidirectional flow can force a pedestrian to get stuck among the incoming people, walk through the congestion, and unintentionally occupy in a dense area, although more comfortable passageways are available. These behaviors are usually produced and simply noticeable. For this reason, the explicit metabolic-energyminimal short-term path planning (MEM) is proposed and added between GPP and LCA to achieve more behavioral fidelity. For energy analysis, the optimal control theory with the objective energy function from the study of biomechanics is employed and finally leads to the very useful optimal walking characteristics for pedestrians. The simulation results show that the pedestrians with MEM can adapt their moving to avoid the congestion, resulting in more promising lane changing and overtaking behaviors. Even though MEM is mainly developed to deal with the artifacts in bidirectional flows, it can be extended with a little modification and produce significant behavioral improvement in multi-directional case.