2020/12/29 by Fahad Islam, Islam, Fahad, Chris Paxton +9 · 2 citations
Computer Science · Engineering · #FOS: Computer and information sciences #Robot Manipulation and Learning #Robotic Path Planning Algorithms #Robotics (cs.RO) #Software Testing and Debugging Techniques
paper · pdf · doi:10.48550/arxiv.2012.14970
openalex publication_date 2020/12/29 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
In many applications, including logistics and manufacturing, robot\nmanipulators operate in semi-structured environments alongside humans or other\nrobots. These environments are largely static, but they may contain some\nmovable obstacles that the robot must avoid. Manipulation tasks in these\napplications are often highly repetitive, but require fast and reliable motion\nplanning capabilities, often under strict time constraints. Existing\npreprocessing-based approaches are beneficial when the environments are\nhighly-structured, but their performance degrades in the presence of movable\nobstacles, since these are not modelled a priori. We propose a novel\npreprocessing-based method called Alternative Paths Planner (APP) that provides\nprovably fixed-time planning guarantees in semi-structured environments. APP\nplans a set of alternative paths offline such that, for any configuration of\nthe movable obstacles, at least one of the paths from this set is\ncollision-free. During online execution, a collision-free path can be looked up\nefficiently within a few microseconds. We evaluate APP on a 7 DoF robot arm in\nsemi-structured domains of varying complexity and demonstrate that APP is\nseveral orders of magnitude faster than state-of-the-art motion planners for\neach domain. We further validate this approach with real-time experiments on a\nrobotic manipulator.\n