Hanjun Song, a masters student from the Korea Advanced Institute of Science and Technology (KAIST), worked on several projects related to manipulation during his internship at Fetch Robotics.
Robots require the ability to plan smooth and predictable motions through space, whether they are manipulating objects or navigating through facilities. A prerequisite of nearly any motion planning algorithm is the ability to check for potential collisions between different objects. This collision checking often consumes a majority of the time spent planning. One of Hanjun’s projects involved developing a new algorithm for creating simple representations of complex objects which allow faster collision checking. Most complex objects are represented by meshes or point clouds, and his algorithm can break these into collections of spheres, which are much faster to check for collisions.
Hanjun’s second project involved adding arm control to our joystick teleoperation. This allows our R&D team and our research customers to more easily move the arm in preparation for a demo or test. Hanjun’s teleoperation work has been released to the ROS buildfarm and will be available on Research-Edition Fetch very soon.