Path Planning of Mobile Robot Based on BiLSTM-D3QN Algorithm

Aiming at the traditional deep reinforcement learning algorithms with slow network convergence, poor reward convergence stability, and low path planning efficiency, this paper proposes a BiLSTM-D3QN path planning algorithm based on the DDQN decision model. Firstly, bidirectional long short-term memory (BiLSTM) is introduced to make the network memorable and increase the stability of decision-making, so that the reward convergence is more stable; secondly, the adaptive reprioritization of experience replay based on frequency penalty function is proposed to extract important and fresh data from the experience pool to accelerate the convergence of the neural network. Lastly, an adaptive action selection mechanism is introduced to optimize the action exploration further. Simulation experiments show that the BiLSTM-D3QN path planning algorithm ha 2 fewer inflection points, 0.52s less planning time, and 4.4% higher success rate than the DDQN algorithm, and outperforms the traditional deep reinforcement learning algorithms in terms of the convergence speed of the network, the efficiency of the planning, and the stability of the convergence of rewards. The superiority of the BiLSTM-D3QN algorithm in network convergence speed and path planning performance is demonstrated.