A deep reinforcement learning based homeostatic system for unmanned position control

Abstract

Deep Reinforcement Learning (DRL) has been proven to be capable of designing an optimal control theory by minimising the error in dynamic systems. However, in many of the real-world operations, the exact behaviour of the environment is unknown. In such environments, random changes cause the system to reach different states for the same action. Hence, application of DRL for unpredictable environments is difficult as the states of the world cannot be known for non-stationary transition and reward functions. In this paper, a mechanism to encapsulate the randomness of the environment is suggested using a novel bio-inspired homeostatic approach based on a hybrid of Receptor Density Algorithm (an artificial immune system based anomaly detection application) and a Plastic Spiking Neuronal model. DRL is then introduced to run in conjunction with the above hybrid model. The system is tested on a vehicle to autonomously re-position in an unpredictable environment. Our results show that the DRL based process control raised the accuracy of the hybrid model by 32%.