KEYWORDS: Mobile robots, Detection and tracking algorithms, Education and training, Control systems, Evolutionary algorithms, Angular velocity, Physics, Computer simulations, 3D modeling, Neural networks
In the field of mobile robot control, the utilization of reinforcement learning methods often faces the challenge of sparse rewards, resulting in suboptimal control performance. This paper proposes an approach that leverages the Diversity is All You Need (DIAYN) framework to dynamically generate reward functions and enhance the policy network weights of reinforcement learning algorithms. A comparative analysis is conducted with traditional reinforcement learning algorithms, namely Deep Deterministic Policy Gradient (DDPG), Deep Q-Network (DQN), and behavior-based robotic and proportional-integral-derivative (BBR PID) planning control algorithm. The results obtained from the CoppeliaSim simulation experiment demonstrate that, under identical training conditions, the DIAYN-based DDPG algorithm exhibits superior learning capability and achieves faster convergence to optimal actions. Consequently, the mobile robot can reach the target point more efficiently and with greater stability.
Visual simultaneous localisation and mapping is a fundamental technology in autonomous mobile robotic systems. The presence of dynamic objects in the environment can lead to incorrect feature matching, and factors such as fluctuating external lighting conditions can introduce instability to the system, thus limiting the practical application of VSLAM. In this paper, a robust VSLAM system for dynamic environments is proposed. Based on ORB-SLAM2, we use YOLO5 to enhance the consistency of the front-end combined with optical flow motion detection to detect dynamic targets in the environment and reject their feature points. Superpoint replaces ORB in the feature extraction process, which further enhances the adaptability of the system to the instability of the external environment. From the experimental results, it can be seen that the improved VSLAM, the real motion trajectory is extremely close to the estimated trajectory, and the error is greatly reduced.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.