We present a grasp detection method that can be used to localize robotic grasp configurations directly from sensor data without estimating object pose. Our method takes a point cloud from an RGBD-camera, like the Microsoft Kinect, as input and produces 6-DOF grasp pose estimates. We use an algorithmic framework that first generates a large number of grasp candidates and then uses machine learning to predict if a candidate is a viable grasp. Our method is evaluated on both challenging tabletop scenarios and an assistive mobile manipulator. Beyond using machine learning for grasp perception, we explore how to extend this approach to pick and place problems where we have to reason about an object’s orientation.