Luis Duffaut

Autonomous and Intelligent Systems Research Laboratory (AIRLab)

Role: Co-Director (with H. Ossareh)

The AIRLab is driving the future of intelligent systems. Our research on data-driven and model-free estimation and control algorithms enable engineered systems to adapt to complex, real-world environments, pushing beyond traditional boundaries. With a dynamic mix of ground robots, aerial drones, robotic arms, and power grid simulators in the lab as our validation platforms, we tackle some of the most exciting challenges in autonomy.

CAREER: A Universal Framework for Safety-Aware Data-Driven Control and Estimation

Role: Principal Investigator

Duration: 2024-2029

Brief Description: This research enables safe data-driven control and estimation methods for autonomy applications. Here we develop a universal framework for the simultaneous design of control policies and safety measures based on recent advances in the mathematical modeling of dynamical systems. The objective is the automatic synthesis of safety-aware control laws for highly complex systems as a function of their real-time input-output information. Specifically, we leverage a novel framework for system representation that systematically encodes its input-output behavior, namely the Chen-Fliess framework, which allows the use of algebraic optimization routines on the system's information, by eliminating the need for a state-space coordinate frame that otherwise will require over-parametrizations that can lead to infeasibility. The specific goals are: (1) provide an algebraic framework for the analysis and optimization of data-driven control systems, (2) develop input-output reachability analysis in the Chen-Fliess framework, and (3) develop data-driven methods for the synthesis of safe control laws based on reachability analysis and control barrier functions in the Chen-Fliess framework.

CREEL: Cold Weather Summit-to-Shore Environmental Observation Network

Role: Task leader - UAS Snowpack Monitoring

Duration: 2022-2025

Brief description: Snowpack measurements are crucial for cold-weather environmental observation. Developing comprehensive sampling regimes that account for the spatial variability of snow depth and area is challenging using traditional methods. This project proposes the deployment of an array of low-cost, distributed sensors for tracking snow depth and coverage together with the characterization of the spatial co-variability of different measurement methods to aid in the design of snowpack measurement networks. Through this effort, we will develop distributed sensing capabilities as well as address the research question: what environmental/terrain factors (e.g., vegetation/canopy structure, micro-topography) enable low-cost tech (imaging, GPS) to be reliable and accurate for snowpack reconstruction? From the methodological perspective, spatial generators based on the self-similar properties of the field of interest will be used to generate higher resolution data sets of snow spatial distribution. Additionally, the team will unveil the signature of LiDAR information to extract features that have the potential for detecting and differentiating snow properties of interest.

CREEL: Army Visual and Tactical Arctic Reconnaissance (AVATAR)

Role: Task Leader - Efficient Real-Time Data Assimilation and Terrain Reconstruction in Cold Regions with Limited Information and Power

Duration: 2020-2023

Brief description: Robust navigation for Arctic missions presents a number of unique challenges, including: lack of network connectivity, limited or unreliable access to satellite positioning systems, relative sparsity of geographic landmarks, limited power, and the highly dynamic nature of the environment. This project seeks to improve robust navigation in the harsh environments by developing a holistic technology for topography and feature reconstruction using sensors implemented on static stations and on a fleet of aerial drones. Our approach explores venues to avoid, recover or overcome GPS jamming, which is a real complication faced by missions, for instance, in Arctic regions. The two key questions that we answer in this project are: how can navigation information be reliably identified in real-time using distributed, sparse, and noisy data from the sensors; and what is the optimal number and configuration of sensors to maximize navigation information. To answer the first question, we propose to integrate three key methodologies: model-free Kalman filter (MKF) for temporal data reconstruction, Kriging for spatial data reconstruction, and set-theoretic methods for uncertainty quantification and robust decision making. For the second question, an efficient optimization routine is proposed to determine the locations where the highest information value to be feed into the estimation algorithms provided by the first question.

NASA-EPSCoR: Algorithms and Test-bed for Agile Attitude Control of Reconnaissance Drones in Atmosphere Lacking or Weightless Environments

Role: Principal Investigator

Duration: 2022-2023

Brief description: NASA’s latest mission to Mars showcased humankind’s ability to deploy exploring aerial drones in outer worlds such as Mars. This was certainly an achievement that gave light into the broader impact of drone technologies currently used on earth for reconnaissance and exploration. Further developing technologies that assist humans in these endeavours has become priority for the US through NASA. The objective of the project was to simulate, built and test drone like autonomous platforms that facilitate risk-less human exploration in places with no atmosphere and/or weightless conditions such as the moon and/or asteroids. Specifically, the proposers will design and build a test-bed for small, agile and efficient drone-like autonomous vehicles that can perform in weightless and/or atmosphere-free environments.

NASA EPSCoR - New Unified Framework for Scalable, Risk-Aware, and Resilient Estimation and Control of Satellite Swarms

Role: Co-Principal Investigator (co-PI)

Main collaborator: H. Ossareh (PI)

Duration: 2021-2023

Brief description: Current satellite swarm missions rely heavily on communications with a ground station. The calculations and final dispatch of most maneuvers and trajectory refinements are calculated on the ground and approved by human operators, and then beamed to the satellites for execution. It is not surprising that this lack of onboard perception and autonomy has limited the type and complexity of swarm missions, especially those beyond the Earth orbit, where communication delays may be prohibitive. To overcome this lack of autonomy and unleash the full potential of satellite swarms, new theories and methods, as well as computationally-efficient and scalable algorithms, in the following key areas are required: (i) uncertainty quantification and robustification; (ii) formation and trajectory planning; (iii) real-time, collision-free navigation and control; and (iv) robust, cooperative estimation and sensor fusion. The proposed project seeks to fill these gaps by researching and developing a validated and fully-integrated computational and operational framework for formation planning, estimation, and control of a satellite swarm, with application to an SAR swarm in close formation.

ARPA-e PlusUp: Packetized Energy Management (PEM) Coordinating Transmission and Distribution

Role: Co-Principal Investigator (co-PI)

Main Collaborator: Mads Almassalkhi (PI)

Duration: 2019-2021

Brief description: This project leverages and advances the bottom-up, demand-side management technology of packetized energy management (PEM) to coordinate Distributed Energy Resources (DERs) at scale to provide synthetic frequency-regulating reserves while simultaneously managing low-voltage grid constraints in real-time. Specifically, the work iin this project extended PEM technology from providing just 5-minute balancing reserves to also provide frequency-regulating reserves like automatic generation control (AGC) and satisfy Synthetic Frequency-Responsive Reserves capabilities.

Collaborative: EAGER: Hybrid Control Architectures Combining Physical Models and Real-time Learning

Role: Principal Investigator

Main Collaborator: W. Steven Gray (Old Dominion University)

Duration: 2018-2020

Brief description: Artificial neural networks have traditionally been the backbone of machine learning. While these biologically inspired learning systems certainly have their strengths, they also have limitations in the context of control engineering. This project developed a new control architecture which combined the advantages of model-based design methods with those of real-time learning. The specific objectives of the project are to (1) advance the theoretical foundations that underpin real-time learning for control applications, including the cascading of these new learning units for deep learning (2) optimize and adapt the novel theoretical results for real-time control of smart grids to provide a priori performance guarantees. The main problem here lies in the uncertainty coming from the over-simplified/poorly modeled dynamics of the grid in addition to the action of renewable resources.

Website: Click here