Research on multi-robot collaborative operation in logistics and warehousing using A3C optimized YOLOv5-PPO model

Introduction

In the field of logistics warehousing robots, collaborative operation and coordinated control have always been challenging issues. Although deep learning and reinforcement learning methods have made some progress in solving these problems, however, current research still has shortcomings. In particular, research on adaptive sensing and real-time decision-making of multi-robot swarms has not yet received sufficient attention.

Methods

To fill this research gap, we propose a YOLOv5-PPO model based on A3C optimization. This model cleverly combines the target detection capabilities of YOLOv5 and the PPO reinforcement learning algorithm, aiming to improve the efficiency and accuracy of collaborative operations among logistics and warehousing robot groups.

Results

Through extensive experimental evaluation on multiple datasets and tasks, the results show that in different scenarios, our model can successfully achieve multi-robot collaborative operation, significantly improve task completion efficiency, and maintain target detection and environment High accuracy of understanding.

Discussion

In addition, our model shows excellent robustness and adaptability and can adapt to dynamic changes in the environment and fluctuations in demand, providing an effective method to solve the collaborative operation problem of logistics warehousing robots.

A Predictable Obstacle Avoidance Model Based on Geometric Configuration of Redundant Manipulators for Motion Planning

When a manipulator works in dynamic environments, it may be affected by obstacles and may cause danger to people around. This requires the manipulator to be able to plan the obstacle avoidance motion in real time. Therefore, the problem solved in this paper is dynamic obstacle avoidance with the whole body of the redundant manipulator. The difficulty of this problem is how to model the manipulator to reflect the motion relationship between the manipulator and the obstacle. In order to describe accurately the occurrence conditions of the collision, we propose the triangular collision plane, a predictable obstacle avoidance model based on the geometric configuration of the manipulator. Based on this model, three cost functions, including the cost of the motion state, the cost of a head-on collision, and the cost of the approach time, are established and regarded as optimization objectives in the inverse kinematics solution of the redundant manipulator combined with the gradient projection method. The simulations and experiments on the redundant manipulator and the comparison with the distance-based obstacle avoidance point method show that our method improves the response speed of the manipulator and the safety of the system.

Multi-Agent Deep Reinforcement Learning for Multi-Robot Applications: A Survey

Deep reinforcement learning has produced many success stories in recent years. Some example fields in which these successes have taken place include mathematics, games, health care, and robotics. In this paper, we are especially interested in multi-agent deep reinforcement learning, where multiple agents present in the environment not only learn from their own experiences but also from each other and its applications in multi-robot systems. In many real-world scenarios, one robot might not be enough to complete the given task on its own, and, therefore, we might need to deploy multiple robots who work together towards a common global objective of finishing the task. Although multi-agent deep reinforcement learning and its applications in multi-robot systems are of tremendous significance from theoretical and applied standpoints, the latest survey in this domain dates to 2004 albeit for traditional learning applications as deep reinforcement learning was not invented. We classify the reviewed papers in our survey primarily based on their multi-robot applications. Our survey also discusses a few challenges that the current research in this domain faces and provides a potential list of future applications involving multi-robot systems that can benefit from advances in multi-agent deep reinforcement learning.

Cooperative collision avoidance in multirobot systems using fuzzy rules and velocity obstacles

Collision avoidance is critical in multirobot systems. Most of the current methods for collision avoidance either require high computation costs (e.g., velocity obstacles and mathematical optimization) or cannot always provide safety guarantees (e.g., learning-based methods). Moreover, they cannot deal with uncertain sensing data and linguistic requirements (e.g., the speed of a robot should not be large when it is near to other robots). Hence, to guarantee real-time collision avoidance and deal with linguistic requirements, a distributed and hybrid motion planning method, named Fuzzy-VO, is proposed for multirobot systems. It contains two basic components: fuzzy rules, which can deal with linguistic requirements and compute motion efficiently, and velocity obstacles (VOs), which can generate collision-free motion effectively. The Fuzzy-VO applies an intruder selection method to mitigate the exponential increase of the number of fuzzy rules. In detail, at any time instant, a robot checks the robots that it may collide with and retrieves the most dangerous robot in each sector based on the predicted collision time; then, the robot generates its velocity in real-time via fuzzy inference and VO-based fine-tuning. At each time instant, a robot only needs to retrieve its neighbors’ current positions and velocities, so the method is fully distributed. Extensive simulations with a different number of robots are carried out to compare the performance of Fuzzy-VO with the conventional fuzzy rule method and the VO-based method from different aspects. The results show that: Compared with the conventional fuzzy rule method, the average success rate of the proposed method can be increased by 306.5%; compared with the VO-based method, the average one-step decision time is reduced by 740.9%.