Incremental Learning of Goal-Directed Actions in a Dynamic Environment by a Robot Using Active Inference

Area of Science:

• Robotics
• Artificial Intelligence
• Cognitive Science

Background:

• Robots need to adapt to dynamic environments.
• Current models struggle with sudden environmental changes.
• Human guidance is often required for robot learning.

Purpose of the Study:

• To develop an active inference-based model for real-time robot adaptation.
• To enable robots to learn incrementally from human tutoring.
• To prevent catastrophic forgetting in robots during continuous learning.

Main Methods:

• Utilized an active inference approach for goal-directed actions.
• Implemented incremental learning from proprioceptive-exteroceptive experiences.
• Incorporated mental rehearsal of past experiences for learning.
• Integrated human tutoring examples for robot guidance.

Main Results:

• The active inference model demonstrated good generalization with optimal parameters.
• Robots showed improved performance on new tasks after few tutoring examples.
• The proposed scheme prevented catastrophic forgetting of previously learned tasks.
• Human intervention was sometimes necessary for sudden, large environmental changes.

Conclusions:

• The active inference model with incremental learning enhances robot adaptability.
• Robots can learn new tasks efficiently from limited human tutoring.
• The approach mitigates catastrophic forgetting, enabling continuous robot learning.