Professor Kolmanovsky's research seeks to synergistically combine advances in control theory with demonstrating the potential of these advances for control applications to aerospace and automotive systems. Recently, along with his students and collaborators, he has been focusing on the following research topics:
Control Theory for Safety Critical Systems with State and Control Constraints for Increased Autonomy
Spacecraft Control
Control for Advanced/Emerging Aerospace Systems
Control of Engines, Powertrain and Propulsion Systems
Autonomous Driving
Control Theory for Safety Critical Systems with State and Control Constraints for Increased Autonomy
Professor Kolmanovsky's research in this area has been advancing methods that are broadly applicable to protecting safety-critical systems across different engineering domains from violating safety constraints during their operation.
As systems are downsized, subjected to more stringent requirements and have to operate more frequently at the performance limits, the ability to non-conservatively handle safety constraints becomes more important. To non-conservatively control systems with constraints (e.g., with actuator, safety, comfort and/or obstacle avoidance limits), controllers must be nonlinear and predictive. Predictive rather than purely reactive control is necessary as current control actions can put the system in danger of constraint violations later. Autonomous systems that operate under stringent state and control safety constraints must be "smart" to optimally use limited resources; they also must be able to learn in dynamically changing environments and do so safely, i.e., without violating constraints.
An example of a control scheme that Professor Kolmanovsky has addressed extensively in his research is the Reference Governor. The Reference Governor is a safety supervisor that monitors reference commands issued by a human operator or by a higher level planning algorithm in an autonomous vehicle. The Reference Governor minimally modifies these reference commands when these commands have a potential to induce constraint violations. The Reference Governor is an add-on solution that augments (rather than replaces) the nominal controller, and has a low computational footprint and simple implementation; hence it is appealing to practitioners. Reference governor theory provides guarantees of constraint satisfaction and finite time convergence of the modified commands to constraint admissible original commands. Extensions of the Reference Governor include Command Governors and Extended Command Governors. While many aspects of reference governor theory have been developed for linear, nonlinear, uncertain, networked, high dimensional and decentralized systems, there are still many open challenges, in particular, in the area of integration of reference governors and learning and achieving fast learning rates while enforcing constraints both during learning and after learning has been completed.
Professor Kolmanovsky has also made contributions to advancing several other approaches to constrained control such as
(i) set-theoretic/multimode controllers which rely on chaining constraint admissible invariant or contractive sets;
(ii) parameter governors (in particular, gain governors, feed-forward governors etc.) for adjusting parameters in nominal control laws online to enforce constraints;
(iii) drift counteraction (viability) maximizing optimal and model predictive control for maximizing time that a system affected by large disturbances/strong drift, that make eventual constraint violation inevitable, can function without violating the imposed constraints;
(iv) action governors which minimally modify control actions to preserve safety of conventional and reinforcement learning based controllers;
(v) constrained control allocation techniques to enforce the prioritized use of the actuators and to prevent Pilot Induced Oscillations (PIO) in aircraft;
(vi) robust controllers that enforce constraints and guarantee steering into the terminal set for systems with set-bounded disturbances;
(viii) optimal control/optimal trajectory characterization in state constrained optimal control problems with applications to numerical methods for constrained spline interpolation and eco-driving.
Professor Kolmanovsky and his students/collaborators have also contributed to advancing Model Predictive Control (MPC) theory for systems with constraints. These contributions include
(i) the development of MPC formulations with guaranteed closed loop properties such as MPC for systems with discrete-time dynamic models evolving on manifolds such as SO(3);
(ii) novel numerical algorithms, approaches and analysis methods for fast and reliable onboard optimization in MPC based on integration of ideas in variational analysis into MPC (including inexact Newton-Kantorovich methods applied to necessary conditions for optimality in MPC in the form of variational inclusions, sensitivity-based warm-starting, semismooth predictor corrector methods, dynamically embedded MPC based on primal-dual gradient flows, and proximally stabilized Fischer-Burmeister (FBstab) quadratic programming algorithm)
(iii) framework for closed-loop stability analysis and estimating required number of iterations for the time distributed optimization (TDO) approach to MPC where a finite number of iterations is applied at each sampling instant,
(iv) supervisory schemes (called Computational Governors and Feasibility Governors) for handling inexact optimization in MPC;
(v) constraint aggregation methods to lower computation complexity in MPC;
(v) constraint separation principle for MPC;
(vi) variable horizon ("Block") contractive MPC with stability guarantees;
(vii) Integrated Perturbation Analysis and Sequential Quadratic Programming (IPA-SQP) solver for nonlinear MPC that exploits neighboring extremal optimal control theory for constrained systems to generate fast predictor-corrector updates;
(viii) methods for hierarchical MPC-based control which exploit lambda-contractive terminal sets.
Professor Kolmanovsky has been pursuing applications and demonstration of constrained control algorithms on a variety of space, air, and ground vehicle platforms. These include automotive engines, fuel cell systems, hybrid electric vehicle energy management, vehicle rollover protection, glider flight management, maneuver and gust load alleviation in Very Flexible Aircraft (VFA), aircraft loss of control prediction and recovery, aircraft gas turbine engine control, coordinated control in more electric aircraft, spacecraft attitude control, spacecraft rendezvous, docking and proximity maneuvering, and flight control of airborne wind energy systems.
Spacecraft Control
Professor Kolmanovsky and his students/collaborators have been demonstrating novel capabilities for constrained control of spacecraft relative motion (rendezvous on Earth and Halo orbits, proximity operations, docking, asteroid landing, etc.) with obstacle/debris avoidance and for spacecraft mission life extension through applications of nonlinear, optimal, constrained and model predictive control. They have been also focusing on methods for constrained spacecraft attitude control, including reaction wheel desaturation; understanding novel pathways to recover controllability and effectively controlling underactuated spacecraft; enhancing observability/parameter identifiability through active maneuvering; feedback control of interplanetary orbit transfers exploiting minimum time Model Predictive Control (MPC) formulation and Lyapunov function-based techniques; spacecraft fault management and reactive mission planning based on stochastic dynamic programming; and applications of learning to safe and non-conservative spacecraft maneuvering. Their publications also address applications of adaptive control and optimal control-inspired time-varying stabilization techniques to spacecraft control problems and procedures to improve robustness and reliability of spacecraft trajectory optimization using indirect methods of optimal control.
In his past research, Professor Kolmanovsky has also considered control of underactuated multibody spacecraft where reorientation and attitude control maneuvers were accomplished using spacecraft shape changes either alone or in combination with the reaction wheels. These control strategies synergistically exploit nonlinear interactions and nonintegrability of the angular momentum exchange for spacecraft attitude control in a similar manner to falling cats and somersaulting athletes. His contributions included the treatment of nonzero angular momentum case and of three dimensional reorientation maneuvers using the theory of averaging.
Control Applications to Advanced/Emerging Aerospace Systems
Professor Kolmanovsky and his students/collaborators have been advancing approaches for maneuver and gust load alleviation to enable safe flight of fuel-efficient flexible and very flexible aircraft, for distributed coordinated control of propulsion and electrical power systems in a more electric aircraft, and for aircraft loss of control mitigation.
Their publications revealed novel capabilities enabled by control theory for airborne wind energy systems and for hydrofoiling catamarans.
Control of Engines and Propulsion Systems
Professor Kolmanovsky has been developing control technology for advanced engines, powertrains and propulsion systems to achieve improvements in fuel/energy efficiency and achieve emission reductions. His more recent publications address control challenges for free-piston engines to achieve piston collision-free operation; constrained and predictive control of turbocharged engines (both diesel and gasoline) to improve their fuel economy, emissions and drivability; control of powertrains with advanced transmissions (dual clutch, CVT); coordinated control of advanced multi-stage hybrid powertrains; integrated control of propulsion and thermal management in connected and automated electric and hybrid electric vehicles; exploiting stochastic control approaches to develop advisory and automated systems for fuel efficient in-traffic driving and vehicle speed control; vehicle to cloud to vehicle control system that fuses off-board and on-board computing resources, and modeling, on-board monitoring and control of container ship marine diesel and dual fuel powerplants. He has contributed to the development of limit protection schemes for aircraft gas turbine engines and for coordinated and distributed control of gas turbine engine and electrical power system in a more electric aircraft. Prof. Kolmanovsky has also been addressing problems in energy management of glider flight and of hybrid powerplants (fuel cell + battery) for small UAV applications.
Autonomous Driving
For Autonomous Driving solutions based on game theory have been pursued to enable autonomous driving in interactive traffic scenarios involving mixed traffic of human-driven and automated vehicles (intersection crossing, highway merging, etc.). Originally these solutions were developed to support V&V and fault discovery for automated vehicles, i.e., of incorrect decisions made by the automated vehicle leading to safety or comfort related issues. They have provided a foundation for the development of an automated vehicle simulator for validation and verification. More recently, these game theoretic approaches have been used to inform several innovative path/decision planning algorithms for automated driving.
Other Research Contributions
Over the years, Professor Kolmanovsky has contributed to several other research topics. These include approaches to stabilization of nonholonomic and underactuated dynamic systems, which is a particularly challenging class of nonlinear control problems. His work has emphasized switching/hybrid control strategies to achieve exponential convergence rates. The results have been both proven theoretically and validated experimentally.
In the area of stochastic control, he and his colleagues have addressed several stability, control and fault estimation problems in systems with random time delays. Such problems have been motivated by issues arising in network control systems.
Several of Professor Kolmanovsky's publications consider control and estimation problems for stochastic systems with jump-diffusion (Wiener-Poisson) disturbances where the jump component represents rare but significant events. He has exploited these techniques for dynamic financial portfolio optimization and energy harvesting.
Another one of his research topics has been the unknown input estimation with the focus on simultaneous input and parameter estimation, for which he and his colleagues have developed and experimentally validated several novel estimation algorithms, and on control based on the idea of using input observer generated estimates for feedforward cancellation. Using insights gained from the treatment of finite dimensional problems, novel algorithms were also developed for source location/intensity estimation in systems modeled by parabolic PDEs, and for boundary conditions estimation in non-local PDEs arising in thermoelasticity.
Professor Kolmanovsky has also considered receding horizon formulations for active state estimation problems, where a control law is designed for simultaneous tracking and estimation accuracy / identifiability enhancement.
.