隨著電子商務與全通路零售的快速發展，倉儲系統逐漸由傳統儲存導向轉型為高即時性與高流動性的訂單履行中心，其中揀貨作業成為影響整體營運績效之關鍵環節。近年來，自主移動機器人（Autonomous Mobile Robot， AMR）與人員協作的揀貨模式逐漸被導入實務環境，以降低人員行走負荷並提升作業效率。然而，在隨機訂單到達環境下，現有研究多聚焦於單一決策層面，對於同時整合訂單批次觸發機制、批次化方式及人機任務選擇策略之聯合決策問題仍缺乏系統性探討；尤其在觸發機制設計中，鮮少針對不同優先級訂單設計差異化觸發條件，使緊急訂單之服務水準難以保障。 本研究以類LocusBots之AMR人機協作揀貨系統為研究情境，建立一套整合「觸發–批次–選擇（Triggering–Batching–Selecting）」決策之動態模擬模型。研究設計四種訂單批次觸發機制（PCT、AOT、HPOT、CSPOT）設置低、中、高三個水準，並比較Seed-order + Accompanying-order Rules與Clustering algorithm Rules兩類批次化策略（合計22種策略組合，涵蓋13種Clustering Rules與9種Seed–Accompanying Rules），透過模擬方法，從系統效率、訂單服務水準與資源利用三個面向，建立六項績效指標，系統性分析不同策略組合對系統效率、訂單服務水準與資源利用等面向之影響。研究結果顯示，Robot與揀貨人員數量組合是決定系統整體表現水準之根本因素，觸發機制與批次化策略則進一步影響資源充足條件下之細部績效；其中，批次大小之選擇是最關鍵的策略槓桿，服務水準與Robot裝載效益之間存在根本性取捨，且此取捨之最優方向會隨系統資源條件完全逆轉。本研究之發現可為AMR人機協作揀貨系統之資源規劃與策略設計提供實證依據。;With the rapid development of e-commerce and omnichannel retailing， warehouse systems have gradually transformed from traditional storage-oriented facilities into highly responsive and dynamic order fulfillment centers， with order picking emerging as a critical factor influencing overall operational performance. In recent years， picking models combining Autonomous Mobile Robots (AMRs) with human collaboration have increasingly been adopted in practice to reduce workers′ walking burden and improve operational efficiency. However， under stochastic order arrival environments， existing research has largely focused on a single decision layer， and systematic investigation into the joint decision problem of simultaneously integrating order batch triggering mechanisms， batching methods， and human–robot task selection strategies remains lacking; in particular， within the design of triggering mechanisms， few studies have designed differentiated triggering conditions for orders of different priority levels， making it difficult to guarantee the service level of urgent orders. This study considers an AMR-assisted human–robot collaborative picking system similar to the LocusBots framework as the research context， and establishes a dynamic simulation model integrating "Triggering–Batching–Selecting" decisions. Four order batch triggering mechanisms (PCT， AOT， HPOT， CSPOT) are designed， each set at low， medium， and high levels， and Seed-order + Accompanying-order Rules are compared against Clustering algorithm Rules across two categories of batching strategies (a total of 22 strategy combinations， comprising 13 Clustering Rules and 9 Seed–Accompanying Rules). Through simulation， six performance indicators are established across three dimensions—system efficiency， order service level， and resource utilization—to systematically analyze the impacts of different strategy combinations on system efficiency， order service level， and resource utilization. The results indicate that the Robot–Picker combination is the fundamental factor determining the system′s overall performance level， while triggering mechanisms and batching strategies further influence fine-grained performance under conditions of sufficient resources; among these， batch size selection is the most critical strategic lever， involving a fundamental trade-off between service level and Robot loading efficiency， with the optimal direction of this trade-off completely reversing depending on system resource conditions. The findings of this study provide empirical evidence to support resource planning and strategy design for AMR-assisted human–robot collaborative picking systems.
