Evolution of Neural Networks for Helicopter Control: Why Modularity Matters (original) (raw)

Real-time evolution of an embedded controller for an autonomous helicopter

2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence), 2008

In this paper we evolve the parameters of a proportional, integral, and derivative (PID) controller for an unstable, complex and nonlinear system. The individuals of the applied genetic algorithm (GA) are evaluated on the actual system rather than on a simulation of it. This makes implicit a formal model identification for the implementation of a simulator. This also calls for the GA to be approached in an unusual way, where we need to consider new aspects not normally present in the usual situations using an unnaturally consistent simulator for fitness evaluation. Although elitism is used in the GAs, no monotonic increase in fitness is exhibited by the algorithm. Instead, we show that the GA's individuals converge towards more robust solutions.

Modelling and Identification of Flight Dynamics in Mini-Helicopters Using Neural Networks

Unmanned Aerial Vehicles have widely demonstrated their utility in military applications. Different vehicle types -airplanes in particular -have been used for surveillance and reconnaissance missions. Civil use of UAVs, as applied to early alert, inspection and aerialimagery systems, among others, is more recent (OSD, 2005). For many of these applications, the most suitable vehicle is the helicopter because it offers a good balance between manoeuvrability and speed, as well as for its hovering capability. A mathematical model of a helicopter's flight dynamics is critical for the development of controllers that enable autonomous flight. Control strategies are first tested within simulators where an accurate identification process guarantees good performance under real conditions. The model, used as a simulator, may also be an excellent output predictor for cases in which data cannot be collected by the embedded system due to malfunction (e.g. transmission delay or lack of signal). With this technology, more robust fail-safe modes are possible. The state of a helicopter is described by its attitude and position and the characteristics of its dynamics system correspond to those of a non-linear, multivariable, highly coupled and unstable system . The identification process can be performed in different ways, on analytical, empirical or hybrid models, each with its advantages and disadvantages. This Chapter describes how to model the dynamic of a mini-helicopter using different kinds of supervised neural networks, an empirical model. Specifically, the networks are used for the identification of both attitude and position of a radio controlled mini helicopter. Different hybrid supervised neural network architectures, as well as different training strategies, will be discussed and compared on different flight stages. The final aim of the identification process is to build a realistic flight model to be incorporated in a flight simulator. Although several neural network-based controllers for UAVs can be found in the literature, there is little work on flight simulator models. Simulators are valuable tools for in-lab testing and experimenting of different control algorithms and techniques for autonomous flight. A model of a helicopter's flight dynamics is critical for the development of good a simulator. Moreover, a model may also be used during flight as predictor for anticipating the behaviour of the helicopter in response to control inputs. The Chapter first focuses on two neural-network architectures that are well suited for the particular case of mini-helicopters, and describes two algorithms for the training of such neural-network models. These architectures can be used for both multi-layer and radial-based www.intechopen.com Aerial Vehicles 602

Neural networks for helicopter azimuth and elevation angles control obtained by cloning processes

2010 IEEE International Conference on Systems, Man and Cybernetics, 2010

Neural networks have been applied very successfully in the identification and control of nonlinear dynamic systems. The paper presents a design of neural network based control system for 2DOF nonlinear laboratory helicopter model (Humusoft CE 150). The main objective of this paper is to develop artificial neural networks to control helicopter's motors, or consequently elevation and azimuth angles. Neural networks are obtained by cloning various type of controllers designed in our previous papers. Those procedures included a cloning linear PID controller, gain scheduling controller and fuzzy controller.

Evolutionary Control of Helicopter Hovering Based on Genetic Programming

Computational intelligence techniques such as neural networks, fuzzy logic, and hybrid neuroevolutionary and neuro-fuzzy methods have been successfully applied to complex control problems in the last two decades. Genetic programming, a field under the umbrella of evolutionary computation, has not been applied to a sufficiently large number of challenging and difficult control problems, in order to check its viability as a general methodology to such problems. Helicopter hovering control is considered a challenging control problem in the literature and has been included in the set of benchmarks of recent reinforcement learning competitions for deriving new intelligent controllers. This chapter shows how genetic programming can be applied for the derivation of controllers in this nonlinear, high dimensional, complex control system. The evolved controllers are compared with a neuroevolutionary approach that won the first position in the 2008 helicopter hovering reinforcement learning competition. The two approaches perform similarly (and in some cases GP performs better than the winner of the competition), even in the case where unknown wind is added to the dynamic system and control is based on structures evolved previously, that is, the evolved controllers have good generalization capability.

Adaptive neural flight control system for helicopter

2009 IEEE Symposium on Computational Intelligence for Security and Defense Applications, 2009

This paper presents an adaptive neural flight control design for helicopters performing nonlinear maneuver. The control strategy uses a neural controller aiding an existing conventional controller. The neural controller uses a real-time learning dynamic radial basis function network, which uses Lyapunov based on-line update rule integrated with the neuron growth criterion. The real-time learning dynamic radial basis function network does not require a priori training and also find a compact network for implementation. The proposed adaptive law provide necessary global stability and better tracking performance. The simulation studies are carried-out using a nonlinear desktop simulation model. The performances of the proposed adaptive control mechanism clearly show that it is very effective when the helicopter is performing nonlinear maneuver.

Evolving modular neural networks to solve challenging control problems

2004

Abstract This article describes ModNet, a framework devoted to the evolution of modular neural controllers that affords possibilities of bootstrapping the search for efficient solutions to challenging problems. Initial knowledge may be provided either as modules assigned to specific computations, or as an overall connectivity pattern describing how modules could be connected to each other or to the controller's inputs and outputs.

Evolution of neurocontrollers for complex systems: alternatives to the incremental approach

2004

Abstract Applications of neural networks to solve challenging control problems still face important difficulties when scalability issues are involved. Usually, such difficulties are tackled according to an incremental approach that consists in decomposing a given task into simpler sub-tasks that may be separately solved. In this article, we describe two alternatives to this approach and demonstrate their efficiency on the control of a simulated lenticular blimp, ie, a complex dynamic platform with five sensors and seven motors.

Neural Network Models for the On-Board and Individual Blade Control of Helicopter Rotors

27th AIAA Applied Aerodynamics Conference, 2009

Neural Networks are evaluated to model On-Board Blade Control concepts for helicopter rotors. Computational Fluid Dynamic simulations of active flap and active twist concepts have been run using the compressible Navier Stokes solver OVERFLOW. Neural Network models were then made as function of Mach number and mean angle of attack for the change in lift, drag, and pitching moment and their associated time constants. These models show the ability to capture the nonlinear effects of stall and shocks. These models are now suitable for incorporation into rotorcraft flight simulation software.

Adaptive Control of a 3 Dof Helicopter Model Using Neural Networks

2007

A Neural Network (NN) combined with PD (proportional-derivative) controller is proposed in this paper for application in underactuated nonlinear systems. The main goal of this work is to solve the reference trajectory tracking problem of a three degrees of freedom (DOF) helicopter platform model with two control inputs obtained by EulerLagrange method. This control technique is derived from the estimate of the helicopter nonlinear function performed by the NN, combined with an outer PD tracking loop and an auxiliary signal that provides robustness in the face of unmodeled bounded disturbances, as well as unstructured unmodeled dynamics. The PD controller design is based on a LQR controller, which is designed by using the helicopter linearized model. Lyapunov second method is employed to establish stable weights adaptation laws, which are tuned on-line, and the control system stability, thereby guaranteeing small tracking errors and bounded control signals. The Lyapunov stability of ...