串結式機器人分散式運動控制之設計與實現

Abstract

串結式機器人具備優越的克服複雜地形能力，於救援、監控、搜索等領域都能有所應用。為使串結式機器人可透過單節之間的合作協調，克服複雜地形，本論文首先提出了分散式行為模式多單元控制，藉由master與slave角色的合作：master角色的攀爬豎板(riser climbing)與登上踏板(tread landing)行為，以及slave角色遵從master的命令，以克服複雜地形。 同時引入螞蟻演算法自主學習，使機器人可在運動過程中，自主調整策略，以增進克服地形的效率。經由螞蟻演算法訓練後，可縮減克服地形時間。並提出序列訓練以改善克服時間過長的問題，提出費洛蒙調整以改善學習反效果問題。 藉由在台階、斜坡上的台階，與階梯三種地形上的實驗結果驗證，經過螞蟻演算法的第一次訓練後，可縮減地形克服時間55%。加入序列訓練後，第一次訓練的地形克服時間標準差，較未加入序列訓練時減少77%。加入費洛蒙調整後，即使經多次訓練，也能保證地形克服時間趨於一致，證明了學習的可靠性。

The chain-type robot possesses great capability to overcome complex terrains so that it can be applied to areas like rescue, surveillance, and exploration, etc. To acheive coordination among single modules for motion on complex terrains, this thesis proposes the behavior-based distributed multi-unit control for the chain-type robot. As the master unit executes either the riser climbing or the tread landing behavior, the slave unit follows the command sent from the master. Via appropriate coordination between the master and the slave, the chain-type robot can thus successfully overcome complex terrains. Moreover, to improve the chain-type robot motion for complex environments, the autonomous learning procedure via the ant colony optimization (ACO) is used. In addition, the sequential training is proposed to decrease the conquering time, and the algorithm of pheromone adjustment is also proposed to suppress the deviated results of the learning. Experimental results on all steps, slope and stairs show that ACO decreases 55% of the terrain conquering time after the first learning procedure; and with the proposed sequential training, the deviation of conquering time for the first learning procedure is 77% less than that without sequential training. With pheromone adjustment, the satisfactory results in deviation of conquering time are guaranteed even after several learning procedures to prove that the present learning is reliable.