TOAST: Trajectory Optimization and Simultaneous Tracking Using Shared Neural Network Dynamics

Abstract

Neural networks have been increasingly employed in Model Predictive Controller (MPC) to control nonlinear dynamic systems. However, MPC still poses a problem that an achievable update rate is insufficient to cope with model uncertainty and external disturbances. In this paper, we present a novel control scheme that can design an optimal tracking controller using the neural network dynamics of the MPC, making it possible to be applied as a plug-and-play extension for any existing model-based feedforward controller. We also describe how our method handles a neural network containing history information, which does not follow a general form of dynamics. The proposed method is evaluated by its performance in classical control benchmarks with external disturbances. We also extend our control framework to be applied in an aggressive autonomous driving task with unknown friction. In all experiments, our method outperformed the compared methods by a large margin. Our controller also showed low control chattering levels, demonstrating that our feedback controller does not interfere with the optimal command of MPC. 1

Shopping with AI: Consumers' perceived autonomy in the age of AI

“…Automatic gear shifting encompasses non-smooth and time-varying dynamics characteristics, hence increasing the lower bound of model bias. We alleviate this problem by providing the history of stateaction pairs to the neural network input so that it can extract contextual information [44,45]. The history length H must be carefully determined, because as H increases, the stateaction space that has to be discovered becomes exponentially larger, making the exploration problem more difficult.…”

“…When the history length exceeds 5, the prediction performance generally decreases. In accordance with prior literature employing the same strategy [44,45,54,61], we determine the history length to be 4.…”

Simulation of Obstacle Avoidance Robots

“…This study only considers static workplace obstructions; moving obstructions caused by moving objects are not included. This study only considered unexpectedly appearing static obstacles, although model-based predictive controller (MBPC) using neural networks and ultrasonic sensors is also utilized to guide mobile robots around unexpectedly appearing static obstacles in their environment [62][63][64][65][66][67][68][69][70][71][72][73][74][75][76][77]. The Dynamic Artificial Neural Network (DANN) approach is used for motion planning for mobile robot paths through [78][79][80].…”

Simulation of Obstacle Avoidance Robots

“…Only static impediments in the robot's workspace are addressed in this study; moving obstacles brought on by moving objects are not taken into account. This study only focused on static obstacles that unexpectedly appeared, but model-based predictive controller (MBPC) using neural networks and ultrasonic sensors is also used to navigate mobile robots around static obstacles that unexpectedly appear in their workspace [62][63][64][65][66][67][68][69][70][71][72][73][74][75][76][77]. Motion planning and mobile robot pathways using the Dynamic Artificial Neural Network (DANN) method [78][79][80].…”