Completion of the Infeasible Actions of Others: Goal Inference by Dynamical Invariant

Area of Science:

• Computational Neuroscience
• Robotics
• Human-Computer Interaction

Background:

• Assisting others requires inferring their goals and intentions.
• Understanding incomplete actions is crucial for effective help.
• Current computational models lack robust intention inference from motor actions.

Purpose of the Study:

• To investigate a computational mechanism for inferring intention and goals from incomplete actions.
• To enable AI to complete actions on behalf of humans.
• To analyze the characteristics of motor control reflecting underlying goals.

Main Methods:

• Analyzed single-link pendulum control tasks with manipulated goals.
• Examined behaviors generated by different goal-oriented tasks.
• Utilized fractal dimension of movements as a characteristic of motor controllers and goals.

Main Results:

• Found that fractal dimension of movements characterizes differences in underlying motor controllers and goals.
• Demonstrated that a simulated pendulum controller can infer the direction of an underlying goal from incomplete actions.
• Showcased the use of fractal dimension as a criterion for movement similarity to complete actions.

Conclusions:

• Fractal dimension of movement trajectories is a viable indicator of inferred goals and intentions.
• This computational mechanism allows for the completion of incomplete actions by AI.
• The findings contribute to developing more intuitive and collaborative human-AI systems.