Sensing expectation enables simultaneous proprioception and contact detection in an intelligent soft continuum robot

Area of Science:

• Robotics
• Biomimetic Engineering
• Control Systems

Background:

• Human brain's perceptual models are key for robotic control in complex interactive tasks.
• Soft robots can leverage similar mechanisms for enhanced environmental interaction.
• Existing methods lack robust real-time perception for soft robot manipulation.

Purpose of the Study:

• To propose and demonstrate an expected-actual perception-action loop for soft robots.
• To enable robust contact detection and deformation source identification in soft continuum robots.
• To enhance the interactive capabilities of soft robots in vision-limited environments.

Main Methods:

• Developed a sensorized soft continuum robot.
• Implemented an expected-actual perception-action loop for real-time feedback.
• Utilized shape sensing to match expected and actual robot configurations.
• Tested contact detection and direction perception under various conditions.

Main Results:

• Achieved rapid contact detection (within 0.4s) with low shape estimation error (1.4%).
• Successfully distinguished between external and internal deformation sources, even when applied simultaneously.
• Demonstrated accurate perception of contact direction (error < 10°) in static and dynamic scenarios.
• Showcased autonomous navigation and human-robot interaction for learning desired configurations.

Conclusions:

• The proposed perception-action loop significantly improves soft robot perception and control.
• This approach enables robust interaction and manipulation for soft robots in complex environments.
• The method has potential applications in autonomous navigation and human-robot collaboration.