Affiliation:
1. Xi’an University of Technology
Abstract
<div class="section abstract"><div class="htmlview paragraph">The traditional Double Deep Q-Network (DDQN) algorithm suffers from slow
convergence and instability when dealing with complex environments. Besides, it
is often susceptible to getting stuck in a local optimal solution and may fail
to discover the optimal strategy. As a result, Unmanned Ground Vehicle (UGV)
cannot search for the optimal path. To address these issues, the study presents
an Improved Dueling Double Deep Q Network (ID3QN) algorithm, which adopts
dynamic ε-greed strategy, priority experience replay (PER) and Dueling DQN
structure. Where, UGV solves the problem of insufficient exploration and
overexploitation according to the dynamic ε-greed strategy. Moreover,
high-priority experience examples are extracted using the priority experience
replay approach. Meanwhile, the Dueling DQN method can effectively manage the
relationship between state values and dominance values. According to the
experiment’s accomplishments, the ID3QN method outperforms the DDQN approach in
terms of stability and rate of convergence, and obtains a better path in UGV
path planning.</div></div>