Brief

Researchers have developed a reinforcement learning framework to control and stabilize a Twin Rotor Aerodynamic System (TRAS). The Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm was employed due to its suitability for continuous state and action spaces, bypassing the need for a system model. Simulation results demonstrated the RL controller's effectiveness, which was further validated against conventional PID controllers under wind disturbances and in real-world laboratory experiments. AI

Researchers have developed a novel control system for quadrotors utilizing a Reinforcement Learning (RL) approach, specifically the Soft Actor-Critic (SAC) algorithm. This method focuses on controlling the quadrotor's thrust vector rather than directly manipulating individual rotor speeds. The RL agent determines the thrust percentage along the z-axis and desired roll and pitch angles, which are then processed by a PID controller to set motor RPMs. This new thrust vector control strategy demonstrates faster training times and achieves smoother, more accurate path-following compared to traditional RPM control methods. AI

Researchers have investigated the impact of dynamic entropy tuning in reinforcement learning for quadcopter control. They compared stochastic policies, which optimize a probability distribution over actions, against deterministic policies that select a single action. The study utilized the Soft Actor-Critic (SAC) algorithm for stochastic policies and Twin Delayed Deep Deterministic Policy Gradient (TD3) for deterministic ones. Findings indicate that dynamic entropy tuning positively influences quadcopter control by mitigating catastrophic forgetting and enhancing exploration efficiency. AI