An action generation model by using time series prediction and its application to robot navigation

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Generating complex behaviors in robots requires sophisticated action generation models.
• Sequential module switching is a common approach for complex robotic tasks.
• Predictive capabilities can potentially optimize action generation.

Purpose of the Study:

• To propose and evaluate a novel action generation model for robots.
• To integrate a prediction unit for anticipating motor primitive module usage.
• To investigate the impact of prediction on robot action generation and navigation.

Main Methods:

• Developed an action generation model composed of multiple motor primitive modules.
• Implemented a prediction unit to forecast the next module in a sequence.
• Tested the model in a simulated robot navigation task.

Main Results:

• The proposed model successfully generated complex actions through sequential module switching.
• The prediction unit demonstrated effectiveness in anticipating module usage.
• The integration of prediction positively influenced the robot's action generation during navigation.

Conclusions:

• The proposed action generation model with a prediction unit is effective for complex robotic behaviors.
• Predictive capabilities enhance the performance of modular action generation systems.
• This approach offers a promising direction for developing more adaptive and efficient robot control systems.