Grant: Discovery Grants Program - Individual

Autonomous robotic systems represent a strategic avenue to protect populations, support health care, and sustain economic prosperity. However, state-of-the-art robotic technologies still face limitations in their sensing and perception capability that limit their actual level of autonomy. Among the challenges encountered, the acquisition of large amounts of data quickly leads to saturation of processing capabilities and compromises educated decision making. This experimental research program focuses on the development of opportunistic, intelligent sensing strategies that combine machine vision and tactile probing to support autonomous decision systems and advanced robotic applications. Opportunistic sensing is defined as intelligently selecting what information to acquire next and from where, strategically based upon a partial understanding of a given environment. The ultimate goal is to minimize the total amount of data required while enhancing decision capabilities of an autonomous system. Opportunistic sensing focuses the collection of information over regions or objects that are of strategic importance for the success, safety, precision and robustness of a robotic operation. The proposed research program builds upon our recent successes in i) opportunistic sensing with RGB-D sensors for industrial applications, ii) optimal task allocation among heterogeneous robotic agents, and iii) visual and tactile sensing for the characterization of deformable objects. The main objective of the proposed research program is to develop integrated perceptual and decisional robotic solutions that combine vision, force and tactile perception with the goal to detect, selectively measure, model and interpret complex environments. Understanding of the scene can support the operation of collaborative mobile robots, or a robot arm/hand assembly to further selectively probe, interact with, and manipulate objects. The objective will be achieved by: i) expanding our formal opportunistic 3D imaging methods to tackle sensors' mobility issues; ii) leveraging state-of-the-art machine learning and visual attention methods to increase performance and robustness for the coordination of distributed robotic agents, and iii) developing formal methods to efficiently acquire and combine visual and tactile information in an efficient framework to support dexterous robotic interaction with deformable objects. This research will contribute fundamental knowledge related to efficient acquisition, representation and decisional methods for robotic applications. It offers the potential for significant breakthroughs in the way knowledge is acquired by unmanned systems, over a range of real-world scenarios where robots can intervene. The technology will be developed and validated on actual sensing and robotic platforms, in various contexts of application including defence/security, collaborative mobile robots, dexterous manipulation of deformable objects, and advanced manufacturing.