工業技術研究院

技術簡介

低溫揀貨作業，人員短缺十分嚴重，而冷藏商品常為較脆弱、多形態不規則的食品或原形蔬果，導致市面沒有揀貨自動化設備，本計畫將以強化學習進行揀貨行為建立，透過Ensemble Learning 快速建立多模態模型，達到自主學習並揀取未學習過商品，突破脆弱、多型態商品無法使用機械手揀貨問題。

Abstract

In low-temperature order picking operations, there is a severe labor shortage. Cold storage goods are often fragile and irregularly shaped items such as perishable foods or unprocessed fruits and vegetables, making it difficult to find suitable automated picking equipment in the market. This project aims to develop intelligent picking behaviors through reinforcement learning, and rapidly build multi-modal models using ensemble learning. The system will enable robotic arms to autonomously learn and pick unfamiliar items, overcoming the current limitations in handling fragile and diverse-shaped products with conventional robotic grippers.

技術規格

■ 商品辨識AI模型(軟體)，主要應用技術包含有： ˙ Object Detection 偵測商品位置 ˙ Image Segmentation 辨識商品類型 ■ 智慧靈巧機械手控制系統(軟/硬體) ˙ 智慧靈巧機械手，同時擁有智慧及力覺控制，有智慧且細心取得商品，升級機械手使用層級，將在更多需細膩、多形狀作業揀取行為、安全加工領域使用。

Technical Specification

Merchandise identification AI model (Software) ˙ Object Detection ˙ Image Segmentation ■ Intelligent Dexterous Robotic Hand Control System (Hardware/Software) ˙ The intelligent and dexterous robotic hand integrates both smart perception and force control, enabling careful and precise object handling. This advancement elevates the functionality of robotic hands, making them applicable to a wider range of tasks that require delicate manipulation, handling of irregularly shaped items, and safe processing in complex operational environments.

技術特色

將力感測器搭接於夾抓處並結合深度攝影機，即時取得目標相對距離，及夾抓在環境接觸過程中之阻抗外力，透過順應性控制之導納控制，得到力追蹤誤差，最後修正參考力道，機器手臂穩定夾取軟、硬質商品。。

應用範圍

物流倉儲業者、商品銷售零售業者

接受技術者具備基礎建議（設備）

機械手臂、攝影機、力覺控制模組、IPC

接受技術者具備基礎建議（專業）

■ 使用者 :機械手臂控制及軟體開發能力 ■ 系統維護者：具備系統開發與維護經驗

聯絡資訊