Neural control and coordination of decentralized transportation robots
MetadatosMostrar el registro completo del ítem
This work presents the modeling, control architecture and simulation of a decentralized multi-robot system for transporting material in a warehouse. Each robot has a task scheduler comprising two different neural networks for task allocation and fault tolerance. The path planner consists of a first-order dynamical state equation to control the robot’s four-wheel asynchronous driving and steering, as well as a partial differential equation to coordinate speeds and arrival times. The task allocation and motion coordination combine the robot’s kinematic control law with a one-layer artificial neural network that classifies five-dimensional symbolic logical equations that define the state transitions between asynchronous events. These events include carry and fetch, material supply, robots stop, obstacle avoidance and battery state. Another multilayer artificial neural network reads the same state inputs for fault detection and recovery. The two neural systems feed forward a navigation planner, which uses a partial differential equation to coordinate the robot’s speed and its relaxation time with respect to the robot in front of it. The energy cost is measured by a Lagrangian function. The proposed planning control scheme was computationally validated through parallel computing simulations. The system is shown to be consistent, reliable and feasible, and it allows for fast navigational tasks.
El ítem tiene asociados los siguientes archivos de licencia: