Towards reasoning and coordinating action in the mental space.

Area of Science:

• Robotics and Artificial Intelligence
• Cognitive Systems and Motor Control
• Computational Neuroscience

Background:

• Cognitive systems require mental simulation capabilities beyond reactive control.
• Action planning involves constraint satisfaction, tool use, and hierarchical goal decomposition.
• Previous systems lacked integrated forward/inverse models for complex motor control.

Purpose of the Study:

• To introduce a novel forward/inverse motor control (FMC/IMC) architecture for cognitive systems.
• To enable mental simulation of actions considering geometric and effort-related constraints.
• To develop a system capable of reasoning, planning, and coordinating reaching and grasping tasks.

Main Methods:

• Developed a FMC/IMC architecture relaxing an internal model to a virtual force field.
• Integrated FMC/IMC with a recurrent neural network (RNN) for goal coordination and sub-goal formation.
• Utilized a 5-DOF robotic arm and stereo vision for real-world task execution and simulation.

Main Results:

• The FMC/IMC system successfully simulated and executed reaching and grasping actions with and without tools.
• The RNN effectively managed goal framing, alternative tool searching, and sub-goal generation.
• Demonstrated non-linguistic reasoning and coordination in a robotic platform solving the 2-stick paradigm.

Conclusions:

• The proposed RNN + FMC/IMC system provides a general solution for cognitive motor control and action simulation.
• The architecture supports flexible adaptation to task constraints and the use of tools.
• This approach advances robotic capabilities in complex task reasoning and coordination, inspired by animal behavior.