Curiosity-Driven Learning for Visual Control of Nonholonomic Mobile Robots

In this paper, we study the problem of visual servoing of nonholonomic mobile robots. Achieving precise positioning becomes particularly challenging within the classical approaches of visual servoing, primarily due to motion and field-of-view constraints. Previous work has demonstrated the effectiveness of deep reinforcement learning in addressing visual servoing tasks for robotic manipulators. In light of this, we propose a novel deep reinforcement learning framework that integrates deep recurrent policies and curiosity-driven learning to tackle the problem of visual servoing of nonholonomic mobile robots. First, we analyze the influence of the nonholonomic constraints on control policy learning, and subsequently, we evaluate our approach on both simulated and real-world environments. Our results demonstrate the superiority of our model in terms of spatial trajectories and convergence accuracy compared to the existing approaches.