Multiobjective Evolution of Biped Robot Gaits Using Advanced Continuous Ant-Colony Optimized Recurrent Neural...

Area of Science:

• Robotics
• Artificial Intelligence
• Optimization Algorithms

Background:

• Biped robots require sophisticated control for stable and efficient locomotion.
• Generating natural and stable gaits is a complex multiobjective optimization challenge.
• Recurrent neural networks (FCRNNs) show promise as central pattern generators for robotic gait.

Purpose of the Study:

• To optimize a fully connected recurrent neural network (FCRNN) for the multiobjective gait generation of a biped robot (NAO).
• To apply a novel multiobjective continuous ant colony optimization (AMO-CACO) algorithm for gait generation.
• To evaluate the generated gait based on walking speed, trajectory straightness, body oscillations, and posture.

Main Methods:

• Formulating gait generation as a constrained multiobjective optimization problem.
• Employing an AMO-CACO-based evolutionary learning approach to solve the optimization problem.
• Utilizing ant-path selection and sampling with accumulated rank for Pareto optimal solution discovery.

Main Results:

• The AMO-CACO algorithm successfully optimized the FCRNN for generating biped robot gaits.
• Simulations demonstrated the effectiveness of the AMO-CACO-based FCRNN compared to other algorithms.
• Pareto optimal FCRNNs were successfully applied to control the gait of a real NAO robot.

Conclusions:

• The proposed AMO-CACO-based FCRNN approach is effective for multiobjective gait generation in biped robots.
• This method provides a robust framework for optimizing complex robotic locomotion tasks.
• The successful real-world application on the NAO robot validates the approach's practical viability.