具機械手臂之履帶式機器人協作任務之實現

Abstract

目前履帶式與機械手臂的相關技術已經越來越成熟，但是大部分的研究，還是將兩者分開來分別進行探討，鮮少討論結合的應用策略，因此本文嘗試結合履帶式機器人的移動導航與機械手臂的物件抓取等功能，以實現跨樓層移動取物、多平台的溝通整合以及具交集工作環境的人機協作任務為目標，提出演算法與系統架構。本文所使用的機器人平台為自行研發裝載了五軸機械手臂的履帶型機器人，透過雷射測距儀和超音波感測器的輔助，搭配牆面校準演算法，完成自動爬梯。為實現近端定位，利用ArUco圖示輔助，引導機器人更精準地移動至目標地，接著使用TensorFlow-Lite提供的物件偵測模型，找出場景物件，並建立3D虛擬環境，再根據場景模型，計算機械手臂的路徑規劃，進行物件抓取。另外本研究透過socket自行開發可以與非機器人作業系統架構開發的機器人進行溝通的簡易方式，讓履帶機器人可以跨平台收到由另一台機器人發送的取物需求，進行跨樓層取物的任務，並透過Mediapipe提供的手勢辨識模型，讓人類使用者以簡易手勢與機器人進行簡易的任務溝通，實現具交集工作環境的人機協作任務。

Currently, the related technologies for tracked robots and robotic arms are more and more mature. However, most current research still studies them separately, rarely discussing the integration strategies for their application. This thesis tries to combine tracked robot's movement navigation and robotic arm object grasping capabilities. The goal is to achieve tasks such as cross-floor object retrieval, communication integration between multiple platforms, and collaborative tasks in a shared workspace. This is done through proposed algorithms and system architecture. The robot platform used in this thesis is a self-developed tracked robot equipped with a five-axis robotic arm, which can climb stairs automatically with the assistance of laser rangefinders and ultrasonic sensors and a wall calibration algorithm. To achieve close-range localization, the robot uses ArUco markers to move more accurately to the target location, and then uses a TensorFlow-Lite object detection model to identify objects in the scene and create a 3D virtual environment. Based on the scene model, the path of the robotic arm is planned to grasp the object.In addition, this study develops a simple communication method using sockets, which allows the robot to communicate with other robots that are not based on the Robot Operating System (ROS) framework. This method enables the tracked robot to receive object retrieval requests from another robot across platforms and perform cross-floor object retrieval tasks. It also uses a Mediapipe gesture recognition model to let human users be able to communicate with the robot through simple gestures, enabling collaborative tasks in a shared workspace.