Multiple Autonomous Vehicles, Perception and Planning
Sensing and reasoning about processes and objects in a large physical environment requires collecting and processing large-scale spatiotemporal data. This is typically done through teams of autonomous agents.
Silvia Ferrari, Mechanical and Aerospace Engineering, and collaborators at Georgia Institute of Technology and Carnegie Mellon University are establishing a unified framework, based on stochastic distributed optimal dual control (SDODC) for decentralized task-aware perception and asynchronous planning for autonomous agents—such as underwater and aerial vehicles. These agents are deployed to collect and process large amounts of complex spatiotemporal streaming data.
The SDODC framework under development is overcoming the technical challenges for decentralized perception and planning in dynamic environment by offering scalability to a large number of agents and compatibility to AI perception algorithms. The framework is based on the novel theory of finite set statistics (FISST), which enables SDODC to offer a unified formalism for perception and planning. SDODC treats the AI perception output and the network of agents as multi-object spatiotemporal densities connected by a spatial map.
Ferrari and her collaborators are establishing mappings between mainstream AI perception algorithms on data streams and density to enable task-aware perception. They are finding the minimum amount of map information to be shared among communicating agents for a satisfactory suboptimal control solution. The researchers are establishing the relationship between the quality of the shared map and the suboptimality of the control solution in order to provide guidance for semantic maps constructed by each agent to be shared with their neighbors.
Ferrari and her collaborators are validating the SDODC theory and algorithms via physical and virtual experiments involving teams of autonomous underwater and aerial vehicles instrumented with vision sensors.