ONRG-NICOP: Distributed high-level scene reasoning with teams of heterogeneous robots
Abstract
Obtaining accurate and comprehensive information from a complex scene is one of the most important and relevant problems in robotics. Yet extremely challenging. When robots are deployed in dynamic environments, they need to be aware of their surroundings to discern friend from foe, or to separate important events that require their immediate attention from those secondary and non-essential. Compared to a single robot acting individually, the optimal combination of information gathered by several robots can yield much better results. Specially when each robot is equipped with different sensors (camera, microphone, laser) and they acquire information from different perspectives at the same time. In this project, we will make scientific advances in high-level scene reasoning with a large team of heterogeneous robots equipped with limited sensing and communication capabilities. Using several robots and heterogeneous sensors introduces novel challenges that need to be addressed for the correct understanding of the scene by every robot. The last years have witnessed substantial progress in the development of distributed algorithms that allow a robotic network to reach agreement over low level features by means of consensus iterations. However, when it comes to high level reasoning and heterogeneous sensing, this problem remains unsolved. In a centralized setup, recognition techniques have proved to be powerful solutions, able to efficiently exploit low-level information from different sensors for abstract reasoning. Thus, the main goal of this project will be to propose novel methods that bring up these two elements together, distributed consensus and high level scene understanding by means of state of the art control and recognition techniques. We will develop algorithms that achieve distributed global perception with a team of heterogeneous robots in time-varying environments. As a motivating example, consider a scenario where several robots are monitoring the traffic in a seaport or a large city event: our aim will be to develop distributed algorithms for a fleet of autonomous heterogeneous robots to (i) globally recognize all the elements (ships, people, cars) and events (departures, arrivals) occurring at the port by appropriate mixing and decision making, (ii) make the optimal decisions, in terms of data acquisition and communication between the robots to keep the scene information as accurate and homogeneous as possible among the fleet. All these actions will be performed using only local onboard sensing in a context of limited communications with nearby robots.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 07, 2019
- Source ID
- N629091912027
Entities
People
- Eduardo Montijano
Organizations
- Office of Naval Research
- United States Navy
- University of Zaragoza