Robotics is at the forefront of technologies that rely on point cloud data produced by 3D sensors, such as stereo, structured light, time-of-light, light detection, and ranging. The 3D representation of the environment provided by these sensors can facilitate robotic tasks such as object detection, pose estimation, motion planning, grasping, and more. In contrast to other data modalities, working with 3D point clouds poses several significant challenges due to the presence of artifacts, non-uniform noise, and variation in density. Robust real-time robotic applications have yet to be reliably achieved due to the complications of processing 3D point cloud data. Moreover, research on 3D point cloud processing has focused on extracting geometric features while ignoring the topological aspects of the data. Understanding the topology of the data is important to overcoming these limitations. This research lays the foundation for understanding the function of topological information obtained from diverse sensor data and advancing applications that are central to robotic perception.
This proposal seeks to enhance the capabilities of perception systems used in robotics applications by developing a framework that can extract topological features from 3D point cloud data. The crux of the proposal relies on using persistent homology to study topological features, such as connected components, holes, and voids, at multiple scales. The research plan includes the design of algorithms and data structures along with robotic perception experiments evaluating these concepts. Specifically, this research will (i) provide fundamental insights into the role of persistent homology for handling 3D point clouds, (ii) design advanced algorithms and data structures for obtaining topological features from 3D sensor data, (iii) determine how topological persistence can complement existing geometric approaches for improving 3D point cloud processing, and (iv) explore new avenues of robotic perception skills that make use of agile, robust, and stable topological data analysis of 3D point clouds. In doing so, this project has the potential to significantly improve the perceptual capabilities of robots. An open source library consisting of software developed for this project will be publicly released for educational and research purposes.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
