Optimizing Robotic Task Sequencing and Trajectory Planning on the Basis of Deep Reinforcement Learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Optimization

Background:

• Traditional robot optimization separates task sequencing and trajectory planning, leading to suboptimal solutions.
• This sequential approach overlooks synergistic effects between these critical robotic problems.

Purpose of the Study:

• To develop a co-optimization model integrating task sequencing and trajectory planning for robots, termed the TSTP problem.
• To propose a deep reinforcement learning (DRL) method for solving the integrated TSTP problem.

Main Methods:

• Formulated the robot task sequencing and trajectory planning problem as a unified TSTP problem.
• Modeled the TSTP optimization as a Markov decision process.
• Developed and applied a deep reinforcement learning (DRL) algorithm to solve the TSTP problem.

Main Results:

• The DRL method achieved 30.54% energy savings over traditional evolutionary algorithms.
• The TSTP model demonstrated an 18.22% energy reduction compared to sequential optimization.
• The DRL approach significantly reduced computational time compared to evolutionary algorithms.

Conclusions:

• The integrated TSTP model effectively addresses limitations of sequential optimization in robotics.
• The proposed DRL method offers a computationally efficient and energy-saving solution for robot task sequencing and trajectory planning.