深度强化学习(DRL)已被成功应用于移动机器人路径规划中,基于DRL的移动机器人路径规划算法适用于高维环境,是实现移动机器人自主学习的重要方法。而训练DRL模型需要大量的环境交互经验,这意味着更高的计算成本。此外,DRL算法的经验池容...深度强化学习(DRL)已被成功应用于移动机器人路径规划中,基于DRL的移动机器人路径规划算法适用于高维环境,是实现移动机器人自主学习的重要方法。而训练DRL模型需要大量的环境交互经验,这意味着更高的计算成本。此外,DRL算法的经验池容量有限,无法确保经验的有效利用。作为类脑计算重要工具之一的脉冲神经网络(Spiking Neural Networks,SNNs)以其独有的生物似真性,能同时融入时空信息,适用于机器人环境感知及控制。结合SNNs、卷积神经网络(CNNs)和策略融合,针对基于DRL的移动机器人路径规划算法进行研究,完成了以下工作:1)提出SCDDPG(SCDDP)算法。该算法利用CNNs对输入状态进行多通道特征提取,利用SNNs对提取的特征进行时空学习。2)在SCDDPG的基础上,提出SC2DDPG(SC2DDPG)算法。SC2DDPG通过设计状态约束策略对机器人运行状态进行约束,避免了不必要的环境探索,提升了SC2DDPG中DRL的收敛速度。3)在SCDDPG的基础上,提出了PFTDDPG(Policy Fusion and Transfer SCDDPG,PFTDDPG)算法。该算法采用分阶控制模式与DRL算法融合,针对环境中的楔形障碍物实施沿墙行走策略,并引入迁移学习对先验知识进行策略迁移。PFTDDPG算法不仅完成了单纯依靠RL不能完成的路径规划任务,还可以得到最优无碰路径。此外PFTDDPG提升了模型的收敛速度和路径规划性能。实验结果证明了所提出的3种路径规划算法的有效性,对比实验结果表明:在SpikeDDPG,SCDDPG,SC2DDPG和PFTDDPG算法中,PFTDDPG算法在路径规划成功率、训练收敛速度、规划路径长度等性能指标上表现最佳。本工作为移动机器人路径规划提出了新思路,丰富了DRL在移动机器人路径规划中的解决方案。展开更多
A novel method of global optimal path planning for mobile robot was proposed based on the improved Dijkstra algorithm and ant system algorithm. This method includes three steps: the first step is adopting the MAKLINK ...A novel method of global optimal path planning for mobile robot was proposed based on the improved Dijkstra algorithm and ant system algorithm. This method includes three steps: the first step is adopting the MAKLINK graph theory to establish the free space model of the mobile robot, the second step is adopting the improved Dijkstra algorithm to find out a sub-optimal collision-free path, and the third step is using the ant system algorithm to adjust and optimize the location of the sub-optimal path so as to generate the global optimal path for the mobile robot. The computer simulation experiment was carried out and the results show that this method is correct and effective. The comparison of the results confirms that the proposed method is better than the hybrid genetic algorithm in the global optimal path planning.展开更多
Ant colony optimization (ACO) algorithm was modified to optimize the global path. In order to simulate the real ant colonies, according to the foraging behavior of ant colonies and the characteristic of food, concepti...Ant colony optimization (ACO) algorithm was modified to optimize the global path. In order to simulate the real ant colonies, according to the foraging behavior of ant colonies and the characteristic of food, conceptions of neighboring area and smell area were presented. The former can ensure the diversity of paths and the latter ensures that each ant can reach the goal. Then the whole path was divided into three parts and ACO was used to search the second part path. When the three parts pathes were adjusted, the final path was found. The valid path and invalid path were defined to ensure the path valid. Finally, the strategies of the pheromone search were applied to search the optimum path. However, when only the pheromone was used to search the optimum path, ACO converges easily. In order to avoid this premature convergence, combining pheromone search and random search, a hybrid ant colony algorithm(HACO) was used to find the optimum path. The comparison between ACO and HACO shows that HACO can be used to find the shortest path.展开更多
Path planning of a mobile robot in the presence of multiple moving obstacles is found to be a complicated problem.A planning algorithm capable of negotiating both static and moving obstacles in an unpredictable(on-lin...Path planning of a mobile robot in the presence of multiple moving obstacles is found to be a complicated problem.A planning algorithm capable of negotiating both static and moving obstacles in an unpredictable(on-line)environment is proposed.The proposed incremental algorithm plans the path by considering the quadrants in which the current positions of obstacles as well as target are situated.Also,the governing equations for the shortest path are derived.The proposed mathematical model describes the motion(satisfying constraints of the mobile robot)along a collision-free path.Further,the algorithm is applicable to dynamic environments with fixed or moving targets.Simulation results show the effectiveness of the proposed algorithm.Comparison of results with the improved artificial potential field(iAPF)algorithm shows that the proposed algorithm yields shorter path length with less computation time.展开更多
文摘深度强化学习(DRL)已被成功应用于移动机器人路径规划中,基于DRL的移动机器人路径规划算法适用于高维环境,是实现移动机器人自主学习的重要方法。而训练DRL模型需要大量的环境交互经验,这意味着更高的计算成本。此外,DRL算法的经验池容量有限,无法确保经验的有效利用。作为类脑计算重要工具之一的脉冲神经网络(Spiking Neural Networks,SNNs)以其独有的生物似真性,能同时融入时空信息,适用于机器人环境感知及控制。结合SNNs、卷积神经网络(CNNs)和策略融合,针对基于DRL的移动机器人路径规划算法进行研究,完成了以下工作:1)提出SCDDPG(SCDDP)算法。该算法利用CNNs对输入状态进行多通道特征提取,利用SNNs对提取的特征进行时空学习。2)在SCDDPG的基础上,提出SC2DDPG(SC2DDPG)算法。SC2DDPG通过设计状态约束策略对机器人运行状态进行约束,避免了不必要的环境探索,提升了SC2DDPG中DRL的收敛速度。3)在SCDDPG的基础上,提出了PFTDDPG(Policy Fusion and Transfer SCDDPG,PFTDDPG)算法。该算法采用分阶控制模式与DRL算法融合,针对环境中的楔形障碍物实施沿墙行走策略,并引入迁移学习对先验知识进行策略迁移。PFTDDPG算法不仅完成了单纯依靠RL不能完成的路径规划任务,还可以得到最优无碰路径。此外PFTDDPG提升了模型的收敛速度和路径规划性能。实验结果证明了所提出的3种路径规划算法的有效性,对比实验结果表明:在SpikeDDPG,SCDDPG,SC2DDPG和PFTDDPG算法中,PFTDDPG算法在路径规划成功率、训练收敛速度、规划路径长度等性能指标上表现最佳。本工作为移动机器人路径规划提出了新思路,丰富了DRL在移动机器人路径规划中的解决方案。
文摘A novel method of global optimal path planning for mobile robot was proposed based on the improved Dijkstra algorithm and ant system algorithm. This method includes three steps: the first step is adopting the MAKLINK graph theory to establish the free space model of the mobile robot, the second step is adopting the improved Dijkstra algorithm to find out a sub-optimal collision-free path, and the third step is using the ant system algorithm to adjust and optimize the location of the sub-optimal path so as to generate the global optimal path for the mobile robot. The computer simulation experiment was carried out and the results show that this method is correct and effective. The comparison of the results confirms that the proposed method is better than the hybrid genetic algorithm in the global optimal path planning.
基金Projects(60234030, 60404021) supported by the National Natural Science Foundation of China
文摘Ant colony optimization (ACO) algorithm was modified to optimize the global path. In order to simulate the real ant colonies, according to the foraging behavior of ant colonies and the characteristic of food, conceptions of neighboring area and smell area were presented. The former can ensure the diversity of paths and the latter ensures that each ant can reach the goal. Then the whole path was divided into three parts and ACO was used to search the second part path. When the three parts pathes were adjusted, the final path was found. The valid path and invalid path were defined to ensure the path valid. Finally, the strategies of the pheromone search were applied to search the optimum path. However, when only the pheromone was used to search the optimum path, ACO converges easily. In order to avoid this premature convergence, combining pheromone search and random search, a hybrid ant colony algorithm(HACO) was used to find the optimum path. The comparison between ACO and HACO shows that HACO can be used to find the shortest path.
文摘Path planning of a mobile robot in the presence of multiple moving obstacles is found to be a complicated problem.A planning algorithm capable of negotiating both static and moving obstacles in an unpredictable(on-line)environment is proposed.The proposed incremental algorithm plans the path by considering the quadrants in which the current positions of obstacles as well as target are situated.Also,the governing equations for the shortest path are derived.The proposed mathematical model describes the motion(satisfying constraints of the mobile robot)along a collision-free path.Further,the algorithm is applicable to dynamic environments with fixed or moving targets.Simulation results show the effectiveness of the proposed algorithm.Comparison of results with the improved artificial potential field(iAPF)algorithm shows that the proposed algorithm yields shorter path length with less computation time.
