Improved Double Deep Q-Network Algorithm Applied to Multi-Dimensional Environment Path Planning of Hexapod Robots.

Area of Science:

• Robotics
• Artificial Intelligence
• Chemical Engineering

Background:

• Detecting pipeline leaks in chemical plants is challenging due to complex, multi-dimensional environments and dynamic conditions.
• Hexapod robots offer superior maneuverability for navigating these complex terrains and multi-level structures.
• Efficient path planning is crucial for hexapod robots to effectively survey and identify potential leak points.

Purpose of the Study:

• To develop an advanced path-planning algorithm for hexapod robots operating in chemical plant environments.
• To address the challenges of identifying transition points and optimizing paths in multi-level, dynamic settings.
• To improve the efficiency and effectiveness of hexapod robots in detecting transportation pipeline leakages.

Main Methods:

• Proposed a novel algorithm: PSO-guided Double Deep Q-Network (PG-DDQN).
• Integrated Particle Swarm Optimization (PSO) to guide the Double Deep Q-Network (DDQN) training process.
• Abstracted the environment into localized maps and conducted comparative experiments against standard DQN and DDQN.

Main Results:

• The PG-DDQN algorithm demonstrated faster convergence compared to standard DQN and DDQN.
• Achieved significant reductions in path-planning time: at least 33.94% over DQN and 42.60% over DDQN.
• Validated the PG-DDQN algorithm through Gazebo simulations and physical experiments on a hexapod robot.

Conclusions:

• The PG-DDQN algorithm substantially enhances hexapod robot mobility and efficiency in complex chemical plant environments.
• This approach provides valuable insights for optimizing path planning to detect transportation pipeline leakages.
• The developed algorithm offers a promising solution for autonomous inspection and maintenance in industrial settings.