Velocity range-based reward shaping technique for effective map-less navigation with LiDAR sensor and deep...

Area of Science:

• Robotics and Artificial Intelligence
• Sensor Technology
• Autonomous Systems

Background:

• Advancements in sensor hardware and artificial intelligence enable robots with human-like sensory and cognitive capabilities.
• These technologies are rapidly transforming various industries, particularly robotics, leading to unprecedented performance levels.

Purpose of the Study:

• To establish an optimal path planning strategy for autonomous driving of mobile robots in unmapped industrial environments.
• To leverage LiDAR sensors and deep reinforcement learning for enhanced robot navigation.

Main Methods:

• Review of mobile robot hardware, LiDAR sensor characteristics, and autonomous driving core technologies.
• Investigation of deep reinforcement learning algorithms, definition of a deep neural network for data conversion, and a reward function for path planning.
• Development and utilization of a simulation environment to test autonomous path planning.

Main Results:

• Successful experimental verification of autonomous path planning in a simulated environment.
• Proposal and validation of an "Velocity Range-based Evaluation Method" to improve performance indicators relevant to real-world applications.

Conclusions:

• The proposed method effectively enables autonomous driving and optimal path planning for mobile robots in complex, unmapped environments.
• The research provides valuable guidance for applying advanced robotics and AI technologies in logistics and manufacturing settings.