Grasp pose detection for control of an assistive robotic manipulator

Robotic Assistive Devices are increasingly on-demand as they improve the quality of life of people, especially with upper limb motor impairments. These assistive devices allow individuals to work independently and perform Activities of Daily Living (ADL) like picking and placing, objects that are impossible to do without the other’s support. The proposed system involves the design of an assistive robotic arm system with 7-Degrees of Freedom (DOF), which could be mounted on the wheelchair. The arm is semi-autonomous and incorporates the user input while executing the planned trajectory. Visual servoing is made use of in tandem with a grasp pose detection algorithm to track and pick up the object of interest. The focus of this research work is to develop a more robust grasp pose detection algorithm using deep reinforcement learning that can detect grasp poses for unseen objects. All modules of the proposed system are designed and tested using the Gazebo simulator.