Control
To exercise directing or restraining influence over a process or system.

Receding Horizon Optimal Control
BAI’s receding horizon optimal (RHO) control combines a user-specified desired response model with a patented method for solving the differential Riccati equations online and in real time. The result is a control law that easily handles multiple effectors, has the robustness properties associated with linear quadratic regulators and emphasizes tracking desired transient response models as opposed to simply achieving a desired steady-state response. BAI engineers have successfully applied RHO to a number of challenging control problems including reconfigurable flight control, control of a helicopter slung load and control of power electronics.

Reference: Ward, D.G., R.L. Barron, J.F. Monaco, Y-J.P. Wei, and T.J. Molnar, "Agile nonlinear receding-horizon optimal control law using neural models of stability and control derivatives," Proc. 1996 NASA High-Angle-of-Attack Technology Conf., Hampton, VA, Sep. 1996.

Feedback Linearization and Inverting Control
BAI has a tremendous amount of experience with inverting flight controllers, having dealt with applications ranging from fixed-wing aircraft and launch vehicles to missiles and precision guided munitions. Inverting control is perhaps the most prominent form of nonlinear control, and has been used extensively for flight control. This technique relies on the use of a nonlinear coordinate transformation to recast the nonlinear system into a linear time invariant form. Linear control techniques may then be applied for control synthesis. Since the control does not have an explicit dependence on time, feedback linearization indirectly provides the benefit of gain scheduling.

Adaptive Guidance and Control
There are essentially two basic types of adaptive control: direct and indirect. The indirect scheme consists of two tasks at each time step. The first is the identification of relevant plant parameters. Then, the controller parameters are computed by solving algebraic equations relating the plant parameters to those of the controller. Direct adaptive control differs fundamentally from its counterpart in that the controller parameters are adapted directly to achieve command tracking and no attempt to identify the plant is made. BAI has extensive experience in the application of both direct and indirect methods for flight control. A number of the algorithms developed by BAI have been validated through flight tests.

Reinforcement Learning and Learning Control
In many applications, a controller must do more than adapt to the current conditions; it must learn and remember a control policy for an unknown plant over time. Reinforcement Learning (RL) is a method that learns an optimal control policy over time given sufficient simulation experience. An advantage of RL is that it can handle complex optimization problems in higher-dimensional space. BAI has been developing methods that extend RL to continuous-time systems and apply it to challenging problems such as automated aircraft recovery and coordinated control of unmanned air vehicles. Other learning control research is focusing on how one can add “memory” to models used in indirect-adaptive controllers by making the onboard models used in the controller spatially local. This approach forms the basis for BAI’s Intelligent Control of UAV’s research.

Reference: Monaco, J.F., D.G. Ward, and A.G. Barto, "Automatic aircraft recovery via reinforcement learning: Initial experiments," Advances in Neural Information Processing Systems 10, (M.I. Jordan, M.J. Kearns, and S.A. Solla, Eds., MIT Press, 1998, pp. 1022-1028.