Intelligent Distributed Sensing Towards Multi-Vehicle Autonomy with Undersea Applications
Abstract
Intelligent Distributed Sensing Towards Multi-Vehicle Autonomy with Undersea ApplicationsMulti-robot coordination has been an acti"ve and attractive field for robotic researchers during the past decade, especially in military and space applications. Past research" has mainly focused on having multiple vehicles fulfill pre-defined missions and often times these teams work beside each other but" not necessarily coordinate their activities. The teams were even less poised to dynamically adapt their individual, as well as thei""r team, behaviors in response to what theyobserve in their environment. Simple solutions to team coordination are accomplished eith"er through a human operator or a single master vehicle that controls the other vehicles. The flaws with these simple approaches are apparent; the human operator makes the system no longer autonomous and is often constrained by delayed or limited sensing of the environment and the single master vehicle introduces the single point of failure.For a team of vehicles to truly operate autonomously" in a coordinated manner, robust distributed team knowledge is needed, which is typically established by frequent exchange of messag""es. By maintaining distributed team knowledge, the vehicles can begin to work together intelligently, but often that is not to accom""plish complex tasks. Most team missions tend to be goal oriented, however, during the execution of these goals a single vehicle or a"" team of vehicles may need toadapt to new roles.As the modality of the autonomous vehicles change from aerial to ground, to surfac""e to underwater vehicles, new challenges are introduced. Some of these challenges include varying bandwidth limitations, a broad ran""ge of sensing capabilities and data/message mediation between the modalities.In this work, we propose creating a distributed archit"ecture that will address two key aspects of multi-vehicle autonomy ~ integration of sensor data into acoustic messaging to aid situa"tional awareness and intelligent multi-sensor data fusion. Consequently, we will also address the impact on sensor fusion when the q""uality of the fused imagery degrades due to poorcommunication. Machine learning methods, such as reinforcement learning, will be in"tegrated for feature extraction and data reduction techniques to circumvent the roadblocks associated with sharing sensor data between vehicles on bandwidth-limited undersea acoustic networks. Machine learning can also help fill the data/knowledge gap when poor communication degrades shared sensor data.This work will leverage previous UMassD/NUWC NURP efforts in a) performing a comprehensive review of frameworks for multi-vehicle cooperation through contextual awareness using machine learning and b) enhancing situational awareness through fusion ofsensor data as well as information via communications with other vehicles. This will be a doctoral program effort that will address many issues at a fundamental level establishing solutions that are comprehensive with extensive testing and implementation.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 01, 2017
- Source ID
- N000141712709
Entities
People
- Ramprasad Balasubramanian
Organizations
- Office of Naval Research
- United States Navy
- University of Massachusetts