Self-Tuning PID Control via a Hybrid Actor-Critic-Based Neural Structure for Quadcopter Control
Authors:
Iman Sharifi,
Aria Alasty
Abstract:
Proportional-Integrator-Derivative (PID) controller is used in a wide range of industrial and experimental processes. There are a couple of offline methods for tuning PID gains. However, due to the uncertainty of model parameters and external disturbances, real systems such as Quadrotors need more robust and reliable PID controllers. In this research, a self-tuning PID controller using a Reinforce…
▽ More
Proportional-Integrator-Derivative (PID) controller is used in a wide range of industrial and experimental processes. There are a couple of offline methods for tuning PID gains. However, due to the uncertainty of model parameters and external disturbances, real systems such as Quadrotors need more robust and reliable PID controllers. In this research, a self-tuning PID controller using a Reinforcement-Learning-based Neural Network for attitude and altitude control of a Quadrotor has been investigated. An Incremental PID, which contains static and dynamic gains, has been considered and only the variable gains have been tuned. To tune dynamic gains, a model-free actor-critic-based hybrid neural structure was used that was able to properly tune PID gains, and also has done the best as an identifier. In both tunning and identification tasks, a Neural Network with two hidden layers and sigmoid activation functions has been learned using Adaptive Momentum (ADAM) optimizer and Back-Propagation (BP) algorithm. This method is online, able to tackle disturbance, and fast in training. In addition to robustness to mass uncertainty and wind gust disturbance, results showed that the proposed method had a better performance when compared to a PID controller with constant gains.
△ Less
Submitted 3 July, 2023;
originally announced July 2023.
Human Leg Motion Tracking by Fusing IMUs and RGB Camera Data Using Extended Kalman Filter
Authors:
Omid Taheri,
Hassan Salarieh,
Aria Alasty
Abstract:
Human motion capture is frequently used to study rehabilitation and clinical problems, as well as to provide realistic animation for the entertainment industry. IMU-based systems, as well as Marker-based motion tracking systems, are the most popular methods to track movement due to their low cost of implementation and lightweight. This paper proposes a quaternion-based Extended Kalman filter appro…
▽ More
Human motion capture is frequently used to study rehabilitation and clinical problems, as well as to provide realistic animation for the entertainment industry. IMU-based systems, as well as Marker-based motion tracking systems, are the most popular methods to track movement due to their low cost of implementation and lightweight. This paper proposes a quaternion-based Extended Kalman filter approach to recover the human leg segments motions with a set of IMU sensors data fused with camera-marker system data. In this paper, an Extended Kalman Filter approach is developed to fuse the data of two IMUs and one RGB camera for human leg motion tracking. Based on the complementary properties of the inertial sensors and camera-marker system, in the introduced new measurement model, the orientation data of the upper leg and the lower leg is updated through three measurement equations. The positioning of the human body is made possible by the tracked position of the pelvis joint by the camera marker system. A mathematical model has been utilized to estimate joints' depth in 2D images. The efficiency of the proposed algorithm is evaluated by an optical motion tracker system.
△ Less
Submitted 7 December, 2020; v1 submitted 1 November, 2020;
originally announced November 2020.
