A Universal Tool Interaction Force Estimation Approach for Robotic Tool Manipulation

Area of Science:

• Robotics
• Control Systems
• Machine Learning

Background:

• Accurate six-degree-of-freedom (6-DoF) interaction forces/torque at the tool-end are crucial for robotic manipulation.
• Indirect measurement via robot wrist sensors faces challenges in real-time decoupling of tool inertial forces, especially with varying tools and grasping postures.

Purpose of the Study:

• To develop a universal approach for estimating tool-end interaction forces applicable to diverse grippers and tools.
• To address uncertainties arising from varying tools, grasping postures, sensor zero-drift, and wrist sensor creep effects.

Main Methods:

• An online-identifiable tool dynamics model based on the Newton-Euler approach was developed for the integrated gripper-tool system.
• Sensor zero-drift was incorporated and identified online within the dynamic model for coarse force estimation.
• A spiking neural network (SNN) was employed to compensate for time-varying wrist sensor creep effects.

Main Results:

• Experimental validation on a robotic arm demonstrated root mean square errors below 0.5 N (force) and 0.03 Nm (torque) for estimated tool-end interaction forces.
• The achieved precision is comparable to in situ measurements.
• The method was successfully applied to robotic scraper manipulation with impedance control, enabling precise and rapid interaction force feedback.

Conclusions:

• The proposed universal estimation approach effectively handles diverse tools and grasping postures.
• The integration of an online dynamics model and SNN provides accurate and robust tool-end force estimation.
• This method enhances robotic manipulation capabilities, particularly in compliant operations requiring precise force feedback.