Sensorless Payload Estimation in an Industrial Robot Using Internal Motor-Current Signals

Area of Science:

• Robotics
• Machine Learning
• Sensor Technology

Background:

• Industrial robots require precise payload estimation for safe and efficient operation.
• Traditional methods often rely on external sensors, adding complexity and cost.
• Motor current signals offer a potential internal, sensorless alternative for payload monitoring.

Purpose of the Study:

• To develop and validate a sensorless method for cycle-level payload mass estimation in industrial robots.
• To investigate the use of internal motor-current signals for payload determination.
• To assess the accuracy and generalizability of the proposed method across different robot units.

Main Methods:

• Acquisition of six-axis current traces from KUKA KR3 robot controllers.
• Statistical feature extraction from current signals.
• Payload regression using a multilayer perceptron (MLP) model.
• Systematic validation using a range of payloads (0.4 kg to 2.6 kg) on two KUKA KR3 R540 manipulators.

Main Results:

• The most accurate MLP configuration achieved a testing mean absolute error (MAE) of 6.75 g.
• Mixed-source training (using data from both robots) reduced the testing MAE to 5.37 g, with 96.88% accuracy within a ±15 g tolerance.
• Direct transfer to another robot resulted in a significant MAE increase to 80.54 g, highlighting a cross-robot generalization gap.

Conclusions:

• Motor-current signals can reliably estimate payload mass during repeated single-axis motion under controlled conditions.
• The sensorless method shows potential for industrial applications such as gripping validation and missing-part detection.
• Further research is needed to address the cross-robot generalization gap for broader applicability.