Achieving natural behavior in a robot using neurally inspired hierarchical perceptual control

Area of Science:

• Robotics
• Biomimetic Engineering
• Control Systems

Background:

• Terrestrial locomotion in legged animals is complex due to high degrees of freedom and unpredictable environments.
• Existing robots often struggle with natural terrain and require extensive training or internal models.
• The vertebrate nervous system achieves locomotion efficiently despite these challenges.

Purpose of the Study:

• To introduce a quadrupedal robot capable of posture control and goal-directed locomotion across uneven terrain.
• To demonstrate a novel control architecture inspired by the vertebrate nervous system.
• To achieve robust locomotion in novel and unpredictable environments without prior training or internal models.

Main Methods:

• Development of a quadrupedal robot.
• Implementation of a hierarchical control architecture using simple negative feedback systems.
• Testing in novel environments with unpredictable disturbances.

Main Results:

• The robot demonstrated robust posture control and goal-directed locomotion on uneven terrain.
• The control system successfully navigated novel environments with unpredictable disturbances.
• The robot performed effectively without relying on internal inverse and forward models or requiring training.

Conclusions:

• A biomimetic hierarchical control system can enable robust terrestrial locomotion in robots.
• This approach offers a promising alternative to traditional robot control methods that rely on complex models and training.
• The developed robot shows potential for applications requiring autonomous navigation in complex natural environments.