Tangle- and contact-free path planning for a tethered mobile robot using deep reinforcement learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Path Planning

Background:

• Tethered mobile robots offer continuous power and communication for extended exploration.
• Cable entanglement with obstacles and the robot poses significant operational risks.
• Existing path planning methods may not adequately address cable-related hazards.

Purpose of the Study:

• To develop a tangle- and contact-free path planning (TCFPP) method for tethered mobile robots.
• To mitigate risks associated with cable snagging and entanglement.
• To enhance the safety and reliability of autonomous exploration using tethered robots.

Main Methods:

• Incorporation of homotopy-aware path planning into deep reinforcement learning.
• Reward function design to penalize cable-obstacle and cable-robot contacts.
• Evaluation across two distinct initial cable configurations: sequential deployment and pre-deployed cable.

Main Results:

• The proposed TCFPP method successfully generates paths that minimize cable contact.
• The approach demonstrates effectiveness in both considered cable deployment scenarios.
• Simulation results show superior contact avoidance compared to naive methods.

Conclusions:

• Deep reinforcement learning with homotopy-aware planning is effective for tangle-free path planning.
• The developed method enhances the operational safety of tethered mobile robots.
• This approach enables more reliable long-duration robotic exploration in complex environments.