Learning Combined Pushing and Grasping Policies on a Mobile Manipulator
For a robot to effectively clean a cluttered area, two basic operations are necessary: a push operation to separate tightly packed objects and a grasp operation to take an object to a new designated area. Research has been done in this area on robotic arms. We propose an extension to this research by implementing the push and grasp techniques in simulation of a mobile manipulator, instead of stationary arms done in previous works. For our project, we train a regular visual pushing and grasping policy in simulation in V-REP with a UR5 robot arm. We then program the Jaco gripper on the MOVO to move in simulation according to the coordinates we give. In the future, we will combine these two to test whether the MOVO can perform visual pushing and grasping policies according to the training.
Read our finding in depth here. Final project in a Grad level class, Topics in Collaborative Robotics. Worked with Kotone Tsuji