Sparse and Precise Reconstruction of Static Obstacles for Real-Time Path Planning in Human-Robot Workspaces

Recent advances in robotics aim at applying the benefits of industrial manipulators to use cases in smaller enterprises and the service sector. In these use cases, no a priori geometry is available for obstacles in the workspace. State-of-art approaches thus employ sensors for a real-time reconstruction of obstacles in the workspace and subsequently avoid collisions through real-time path planning. Usually, reconstruction exposes obstacles to path planning over a copious data structure (e.g. an occupancy grid). This data structure, however, is not a good fit for common static and piece-wise planar obstacles (e.g. tables, shelves). We therefore contribute a novel, three-part approach: At first, we create an a priori, sparse and precise reconstruction of static, piece-wise planar obstacles in form of a boundary representation through a hand-held depth sensor. Subsequently, we combine our sparse reconstruction with an online reconstruction of dynamic obstacles to finally enable online path planning. Our experiments show two main benefits of our contribution: Improved reconstruction precision and reduced execution times for collision queries in the path planner. Both of these allow us to increase the motion speed of the manipulator, for empty and for occupied workspaces.