An aerial robot for rice farm quality inspection with type-2 fuzzy neural networks tuned by particle swarm optimization-sliding mode control hybrid algorithm

Agricultural robots, or agrobots, have been increasingly adopted in every aspect of farming from surveillance to fruit harvesting in order to improve the overall productivity over the last few decades. Motivated by the compelling growth of the agricultural robots in modern farms, in this work, an autonomous quality inspection over rice farms is proposed by employing quadcopters. Real-time control of these vehicles, however, is still challenging as they exhibit a highly nonlinear behavior especially for agile maneuvers. What is more, these vehicles have to operate under uncertain working conditions such as wind and gust disturbances as well as positioning errors caused by inertial measurement units and global positioning system. To handle these difficulties, as a model-free and learning control algorithm, type-2 fuzzy neural networks (T2-FNNs) are designed for the control of a quadcopter. The novel particle swarm optimization-sliding mode control (PSO-SMC) theory-based hybrid algorithm is proposed for the training of the T2-FNNs. In particular, the continuous version of PSO is adopted for the identification of the antecedent part of the T2-FNNs while the SMC-based update rules are utilized for the online learning of the consequent part during control. In the virtual environment, the quadcopter is expected to perform an autonomous flight including agile maneuvers such as steep turning and sudden altitude changes over a rice terrace farm in Longsheng, China. The simulation results for the T2-FNNs are compared with the outcome of conventional proportional-derivative (PD) controllers for different case studies. The results show that our method decreases the trajectory tracking integral squared error by %26 over PD controllers in the ideal case, while this ratio goes up to %95 under uncertain working conditions.