近年以來，多足機器人的運動學已被廣泛研究。該研究領域的主 要問題在於環境的不確定性。為了解決這個問題，本研究提出了一 種稱為Adaptive Submodularity with Hypothesis Pruning(ASHP)的方 法， 將不規則地形上的運動問題重新定義為覆蓋Perlin Noise領域中的 問題。多足機器人能夠在沒有外部感受器的情況下，漸漸地適應在復 雜地形上的運動。Adaptive Submodularity在本文中被應用於預測地 形的樣貌，並提供相關的理論保證值。模擬和實驗顯示出本文提出的 方法相比於Random Selection有更小的預測誤差，且相較於其他的模 型(RMA)有更高的成功率及穩定性。;The locomotion of the legged robot has been widely researched in recent years. The main issue in this research area is environmental uncertainty. To overcome this issue, this research proposed a method called Adaptive Submodularity with Hypothesis Pruning(ASHP), which reformulates the locomotion problem on irregular terrains as the coverage problem in the Perlin domain. The legged robot is able to adaptively select locomotion over complex terrains without exteroceptive sensors. The adaptive submodularity is utilized to predict the terrain with theoretical guarantees. The simulations and experiments demonstrate that the proposed approach has less prediction error and a higher success rate than the benchmark, the experiments also show that ASHP is more robust than the benchmark method(RMA).
