Team members: Guoqing Huang , Benyun Zhao , Zimeng Guo , Ziheng Huang

This is our MAEG5755 Project,our team spend one month doing it

First,we know camera get the depth image and will sent to the Dex-Net 4.0 to get the grasp candidates.Then combine with inverse kinematics, the most approrpiate candidate is selected as the grasp pose.at last, run the robot by using inverse kinematics to achieve the pick and place task.

roslaunch deep_grasp_task baxter_simulation.launch

roslaunch moveit_task_constructor_dexnet dexnet_baxter_demo.launch

./baxter.sh
roslaunch moveit_task_constructor_dexnet dexnet_baxter.launch 

calibration we follow chenwei and I will put it on the internet. The network question follow Kenson

• dex-net 4.0
• moveit support packages
• baxter related packages
• baxter_pykdl

simulation: https://www.bilibili.com/video/BV1kq4y1E72k/ https://www.bilibili.com/video/BV1ag41137jW/

real robot: https://www.bilibili.com/video/BV14q4y1E7Ds/

• Ubuntu 20.04 LTS
• ROS : Noetic
• NVIDIA : 460
• CUDA : 10.0
• CUDNN : 7.4
• Tensorflow_gpu-1.13.1

we need to add a lot dependence files in our system , such as python3.6 , due to we use the 20.04 and we may have python3.8 , and we should get the kdl follow the http://github.com/orocos/orocos_kinematics_dynamics . and some dependences we need follow the https://github.com/PickNikRobotics/deep_grasp_demo#Install-Dex-Net to set. and some packages in https://github.com/mfkenson/MAEG5755-2021-Team-PARK. and https://github.com/robot-chenwei/MAEG5755_Robotics_Project.