Directed Morphological Evolution for Fast Robot Structure Optimization
提出定向形态演化方法，通过检测适应不良并利用沙普利值引导突变，加速机器人结构优化，在仿真中使最终适应度翻倍、搜索时间减半。
In biological systems, organisms undergo continuous, intertwined processes of evolution and learning, embedding intelligence in morphology that supports robust and adaptive behaviors in complex environments. Inspired by this principle, evolutionary robotics explores the co-evolution of body and control, transferring part of the computational burden from the controller to the physical body. However, the co-evolution paradigm faces two key challenges. The vast, high-dimensional morphological space that dilutes selective pressure, leading to suboptimal morphologies, and the high computational cost of lifetime-fitness evaluation, which slows evolutionary feedback. Here, we introduce Directed Morphological Evolution (DME), which integrates plasticity-inspired morphology-environment adaptation feedback with a Shapley-guided mutation matrix. Upon detecting maladaptation, DME promptly terminates the current evolutionary lineage and then directs the search toward higher-contribution mutations based on their relative contributions. In this way, DME narrows the exploration to the most promising regions of the search space, eliminating wasteful evaluations. At the same time, DME provides interpretable evolutionary trajectories, where each morphological mutation is guided by a quantified contribution and triggered by environmental maladaptation or prolonged performance stagnation. In simulation experiments across four increasingly complex terrains, DME consistently outperforms baseline methods, approximately doubling final fitness and halving search time. Furthermore, the contribution analysis shows that a small subset of key mutation operators accounts for roughly 85% of cumulative contributions, validating DME’s ability to identify and exploit the key mutation operations, and revealing that different environmental conditions lead to divergent evolutionary trajectories.