CAREER: Leveraging Collective Power of Robotic Grasping Algorithms via Meta-Learning and Active Perception

For many manufacturing, logistics, and service applications, robots are required to pick up and manipulate objects. Even though this fundamental capability has been studied extensively in the last decades and a significant progress has been made, robots still struggle to reach the desired reliability levels, especially when they attempt to manipulate objects in cluttered and unstructured settings. In these settings, the variety of objects and the possible scene configurations are immense, making it extremely challenging to develop a single overarching method that can work in all the conditions. Instead of trying to develop a panacea, this Faculty Early Career Development (CAREER) project presents a fundamentally different approach: leveraging the capabilities of multiple different methods, combining their strengths and avoiding their drawbacks. The framework also creates conditions that boost the algorithms’ success by allowing the robot to efficiently collect more information about the scene. The outcomes of this research will be utilized to develop robotics solutions to environmental problems, e.g. waste sorting and recycling, and establish a first-of-its-kind environmental robotics undergraduate track. For combining the opinions of different algorithms, several ensemble learning methods will be developed, tailored to the robotic manipulation domain. A diversity analysis will be conducted, which will identify the differences between the algorithms and guide the ensemble development process. For enabling the robot to systematically collect data, active vision strategies will be developed for the underlying grasping algorithms and their ensembles. A study to develop a high-level decision-making algorithm is also planned to enable robots to determine the best suited dexterous picking strategy for a given manipulation scenario. 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.