• autonomous picking
• Baxter
• collaborative robotics
• computer vision
• Kinect v2
• Moveit!
• path planning
• ROS
• Universal Jamming Gripper

The work presents an autonomous collaborative robotics system for pick and place of objects in cluttered environments leveraging on Baxter robot as cobotics platform. One of the robot's arm is equipped with a traditional two-fingered gripper, while the second one embeds a custom Universal Jamming Gripper. The motion of the latest one is determined by a custom depth-based perception algorithm which can identify the grasping point for unknown objects. The other arm with the parallel-jaw gripper is controlled through a state of the art mechanism based on point clouds which searches for antipodal grasping points. The motion of both arms is planned using the Moveit! framework. It allows to bring the arms to the picking point and to place the object into a chest avoiding collisions with the environment. Furthermore, the custom gripper with its vision system is compared with two state of the art systems: the work of the MIT-Princeton team that won the stowing task of the Amazon Picking Challenge in 2017 and the system used on the other Baxter's arm. The comparisons show that the proposed system is competitive for the grasping success rate while outperforms the other systems for the searching time of the grasping points.