Multi-Agent Systems Engineering
The current process of developing, testing, and fielding collaborative autonomous systems is time consuming, costly, and error prone, involving unsustainable amounts of human labor, custom tool development, machine-specific integration, one-off component acquisition, manual testing and debugging, and more. Moreover, testing that provides evidence about the safety of collaborative autonomous systems and assessments of their expected performance is essential for widespread adoption and gaining the trust of human operators. ArtIAMAS will address the community’s lack of standard shared test infrastructures, baseline scenarios, tools, common models, etc. and poor interoperability between existing models and tools to enable seamless development of collaborative autonomy across its life cycle. This multi-stakeholder, networked resource will support the rapid conceptualization, development, testing, and deployment of collaborative autonomy.
Enabling Collaborative RDT&E through Shared Infrastructure
This project will focus on developing tools, methods, and processes that support simulation of multi-domain operations that include unmanned ground vehicles that are running the ARL ground autonomy stack and unmanned aerial vehicles (UAVs) that are running the ARL aerial autonomy stack. Anticipated Outcomes: tools, methods, and processes for running simulation experiments of multi-domain operations with UGVs and UAVs; and tools, methods, and processes for running field experiments with autonomous UGVs.
Perception and Robot Navigation of Complex Terrains
\This project will develop new technologies for perception and planning and integrate them with the ARL ground autonomy stack for autonomous navigation. Anticipated Outcomes: methods that combine sparse aerial data with dense ground sensors; improved machine learning methods that can analyze and segment complex large terrains; new machine learning methods to navigate complex robots using multi-sensor perception output; autonomous navigation algorithms for legged robots in complex terrains; methods that integrate uncertain perception and navigation algorithms; and evaluation of robot perception and navigation algorithms on outdoor terrains at R2C2.
Simulation-Based Verification for Autonomous Systems
This project will develop and evaluate simulation-based methods for building assurance cases and verifying that system performance is robust to environmental uncertainties and adversarial actions; it will yield a set of tools enabling practical simulation-based verification of autonomous systems that support the construction of quantified assurance arguments for performance and robustness. Anticipated Outcomes: a rigorous process for specifying assurance cases for autonomous systems; templates for building assurance cases; algorithms for the design of experiments and simulations; data collection and analysis to provide evidence for assurance cases; and scenarios incorporating autonomous systems and an intelligent adversary.
Applied Aerial Autonomy for Long Distance Collaborative Operations
This project seeks to extend multi-vehicle flight systems for operations beyond visual line of sight including cooperative aerial search for ground targets. Anticipated Outcomes: a decentralized search planning and detection disambiguation framework for camera-based target detection with multiple swarm agents; incorporation of the search planning and target detection framework within ARL’s aerial autonomy stack, including simulated and experimental testing at R2C2; integration of perception frameworks including gesture recognition and target reacquisition; and a framework to enable coordination with wheeled or legged ground vehicles that includes air-to-ground guidance, navigation, and control.
Distributed Robotic Beamforming
This project will investigate distributed beamforming with mobile agents to provide continuous, high-power, and reliable signals to mobile nodes while simultaneously guaranteeing low power signals at adversarial locations. Anticipated Outcomes: algorithms for distributed beamforming for a single mobile receiver while null forming at a target location; algorithms for distributed beamforming for multiple mobile nodes while null forming at multiple target locations; and integration of the developed algorithms into the ARL ground autonomy stack for experimental validation.
Risk-aware Heterogeneous Multi-Robot Planning and Coordination for Long-Term Persistent Monitoring of Uncertain, Dynamic Environments
This project will develop risk-aware route planning and coordination algorithms that can reason about uncertainty in a provable fashion to enable long-term autonomous deployments in challenging conditions. Anticipated Outcomes: scalable, risk-aware route planning and coordination algorithms for a team of unmanned air and ground vehicles for long-term missions in partially known, uncertain environments; system analysis on how uncertainty in energy, position, team composition, and system parameters influence performance; novel learning algorithms for single and multi-agent settings with strong theoretical guarantees; novel algorithms with theoretical communication complexity bounds; and simulated and field experiments with aerial and ground robots.
Quick Path Replanning in Response to Opportunistic Multi-Robot Information Gathering about Hazards/Adversaries
This project seeks to enable distributed multi-robot to combine opportunistic information gathering about hazards and adversaries with path-replanning algorithms. Anticipated Outcomes: prototype algorithms and data structures for information-based replanning in simulation; tests and experiments of algorithms and data structures for information-based replanning in simulation; and implementation of algorithms and data structures on hardware platforms.
Cooperative Re-Planning via Sensor Inference and Geospatial Data Integration
This project seeks to enable a heterogenous team of unmanned aerial vehicles to autonomously deploy sensors into a partially known environment using information contained in an ARL-developed Geospatial Data Integration Server (GDIS) to determine deployment locations. Anticipated Outcomes: a sensor placement algorithm modified to work within the mission specifications and adapted to include pre-existing geospatial data; a customized robot interface to GDIS that allows for visualization of terrain and sensors; an adaptive sensor placement algorithm that utilizes updates in geospatial data to iteratively improve on environmental awareness; and simulated and experimental demonstration of algorithms and mission using the ARL aerial autonomy stack.
Multi-Agent Aerial Command and Control via Smart Binoculars
This project seeks to create a line-of-sight smart-binocular system that allows an operator to designate the desired search location of an aerial swarm. Anticipated Outcomes: a framework for interacting with and controlling multi-agent systems using smart binoculars with laser range finding and voice control; and the capability of an operator to specify the location and size of an area to be searched by an aerial swarm.