Grasp detection #4
Description
Input: point cloud
Output: gripper 6d pose where closing the gripper results in a successful grasp
Eventually we'll need to make intelligent guesses about how to pick up known and unknown objects. This topic came up during today's meeting and both agile_grasp and fetch_grasp_suggestion. Also worth a mention is the third iteration on agile grasp, gpd. We never had much success configuring gpd when we tested it for BWI and Villa.
agile_grasp worked well for BWI's needs, but we had much more predictable performance with heuristic grasps (like, just aim for the centroid). Good heuristics coupled with servo-ing proved adequate for Villa.