A brain-inspired intention prediction model and its applications to humanoid robot

Area of Science:

• Robotics
• Artificial Intelligence
• Neuroscience

Background:

• Current human-robot interaction relies on pre-programmed commands, lacking flexibility and often leading to poor user experience.
• Adapting robots to user habits is crucial for seamless integration into daily life and improving user satisfaction.

Purpose of the Study:

• To develop an adaptable, simple, and flexible human-robot interaction technology.
• To enable robots to predict and act according to user intentions using a brain-inspired model.

Main Methods:

• Proposed a brain-inspired intention prediction model based on the neural mechanism of reinforcement learning.
• Utilized spike-timing-dependent plasticity (STDP) mechanisms and binary feedback (right/wrong) for model training.
• Tested the model on a humanoid robot (NAO) in intention prediction and trajectory tracking experiments.

Main Results:

• The proposed model successfully predicted user intentions in experimental settings.
• Demonstrated a significant reduction in training times compared to the traditional Q-learning method.
• The reduction in training times is quantified as (N^2 - N)/4, where N represents the number of intentions.

Conclusions:

• The brain-inspired intention prediction model offers a more flexible and user-centric approach to human-robot interaction.
• This method enhances robot adaptability by learning user intentions effectively.
• The model shows potential for improving user experience in service robots and other applications.