基於柔性行動者評論家之雙臂機器人的碰撞避免

本論文針對雙臂機器人，提出一個基於柔性行動者評論家之碰撞避免的方法和一個訓練方法。主要有兩個部分：(1) 運動控制以及(2) 基於柔性行動者評論家之碰撞避免。在運動控制部分，本論文透過一組虛擬之三連桿的假設，藉由正運動學與幾何方法來獲得七自由度冗餘手臂之逆運動學的解。在基於柔性行動者評論家之碰撞避免的部分，雙臂機器人之左臂與右臂是各由一個神經網路來控制末端點之移動向量和姿態。本論文首先在Gazebo模擬器中建構了一個3D動態模擬環境來作為神經網路之訓練環境，使用Gazebo模擬器之感測器來偵測機器人與環境之距離，以及使用一個基於連桿資訊的偵測方式來避免機器人自身之碰撞。由於直接訓練兩個手臂之神經網路會讓訓練環境過於複雜，因此本論文建立兩個執行緒來分別訓練左臂與右臂之神經網路，並且將其中一個手臂視為另一個手臂之環境物件來降低訓練過程中的複雜度。此外，本論文選擇一些合適之神經網路輸入以及設計一些獎勵函式來讓所訓練完成的神經網路可以控制雙臂機器人，使其在任務執行的過程中能夠有效地避免碰撞。

關鍵字

雙臂機器人；冗餘機械手臂；碰撞避免；深度強化學習；柔性行動者評論家

並列摘要

In this thesis, a collision avoidance method and a training method based on soft actor-critic are proposed for a dual-arm robot. There are two main parts: (1) motion control and (2) collision avoidance based on soft actor-critic. In the motion control, a set of virtual three-link assumptions is used to obtain the inverse kinematics solution of the seven-degree-of-freedom redundant robot manipulator through the positive kinematics and geometric methods. In the collision avoidance based on soft actor-critic, the left-arm and right-arm of the dual-arm robot are each controlled by a neural network to control the movement vector and posture of the end point. A 3D dynamic simulation environment is first constructed on the Gazebo simulator as the training environment of the neural networks. The sensor of the Gazebo simulator is used to obtain the distances between the robot and the environment and a detection method based on link information is used to avoid the collision of the robot itself and the environment. Because directly training the neural networks of the two arms will make the training environment too complicated, two threads are established to train the neural networks of the left-arm and the right-arm separately, and one arm is treated as an environmental object of the other arm to reduce the complexity of the training process. In addition, some appropriate neural network inputs are selected and some reward functions are designed to let the trained neural networks can control the dual-arm robot to effectively avoid collisions during the task execution.

並列關鍵字

Dual-Arm Robot ； Redundant Robot Manipulator ； Collision Avoidance ； Deep Reinforcement Learning ； Soft Actor-Critic

參考文獻

[1] J.F. Engelberger, Robotics in Service, MIT Press, 1989.

Google Scholar

[2] Rethink Robotics Baxter, URL: https://www.generationrobots.com/en/401514-baxter-robot-research-prototype.html

Google Scholar

[3] A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao. “Yolov4: Optimal speed and accuracy of object detection,” arXiv preprint arXiv:2004.10934, 2020

Google Scholar

[4] J. Gehring, M. Auli, D. Grangier, D. Yarats, and Y.N. Dauphin. “Convolutional sequence to sequence learning,” arXiv preprint arXiv:1705.03122v2, 2017.

Google Scholar

[5] E. Wu and H. Koike. “FuturePose - Mixed reality martial arts training using real-time 3D human pose forecasting with a RGB camera,” IEEE Winter Conference on Applications of Computer Vision (WACV), pp.1384–1392, 2019

Google Scholar

國際替代計量

基於柔性行動者評論家之雙臂機器人的碰撞避免

全文下載

主題瀏覽