Milking Robot Arm Control With ANN Feedforward Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Milking robots face challenges in controlling robotic arms due to uncertain and non-linear hydraulic actuators, which are difficult to model, leading to inefficiencies in movement and precision tasks such as attaching teat-cups and cleaning, while existing control solutions like PID regulators are inadequate.

Innovation Solution

A controller incorporating a feedforward module with a trained artificial neural network (ANN), specifically a recurrent neural network (RNN), is used to generate predicted control signals, combined with a closed-loop controller and summation modules to enhance trajectory tracking and position control of the robotic arm.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a closed-loop regulator (PID control) is used to control the robotic arm, then the system can operate with feedback correction, but the trajectory tracking accuracy is insufficient due to non-linear hydraulic actuators and uncertain variables

Engineering Contradiction:
Improvetrajectory tracking accuracyVSAvoidcontrol precision under non-linear conditions
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent applies feedforward control that calculates and applies compensation signals before the tracking error occurs. The controller pre-computes the control actions needed to counteract known disturbances and non-linear effects, allowing the system to proactively maintain accuracy rather than reactively correcting errors after they occur.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent combines feedforward control with feedback correction in a hybrid control architecture. The feedback component continuously monitors actual tracking performance and adjusts the control signal to eliminate residual errors, while the feedforward component handles the dominant non-linear effects, creating a synergistic control system that achieves high precision.

Inventive Principle:
Principle #23Feedback

2Ease of manufacture

If traditional PID control is used, then the control system is simple to implement, but it cannot accurately handle the non-linear relationship between control current and joint movement

Engineering Contradiction:
Improvecontrol system implementation simplicityVSAvoidjoint position accuracy
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent introduces an intermediate compensation calculation layer between the simple PID controller and the hydraulic actuator. This intermediary component computes feedforward compensation signals based on the desired trajectory and system characteristics, translating simple control commands into precise actuator commands that account for non-linearities without requiring complete redesign of the control architecture.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent dynamically adjusts control parameters including feedforward compensation magnitudes and feedback gains based on operating conditions and desired trajectory characteristics. This allows the controller to adapt to different non-linear regimes and optimize performance for various movement scenarios while maintaining a relatively simple overall system structure.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If feedforward compensation is added to improve trajectory tracking, then accuracy improves, but the device complexity increases

Engineering Contradiction:
Improvevelocity trajectory accuracyVSAvoidcontroller structure complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the control function into distinct segments: a feedforward compensation module that handles trajectory prediction and non-linear compensation, and a feedback module that handles error correction. This segmentation allows each module to be optimized independently and simplifies the overall control architecture by clearly separating predictive and reactive control functions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent designs the feedforward compensation mechanism to handle multiple types of disturbances and non-linear effects through a unified calculation framework. The same compensation structure can address gravitational effects, friction, inertial forces, and hydraulic non-linearities simultaneously, reducing the need for separate specialized controllers for each effect and thereby limiting complexity growth.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4656334A1Milking robot controller, method therefore, computer program and non-volatile data carrier
Publication Date: 2025.12.03 DELAVAL HLDG AB
  • EP4656334A1 patent drawingFigure 1~3
  • EP4656334A1 patent drawingFigure 4~5
  • EP4656334A1 patent drawing

AI summary

A controller (100) controls an end-effector (390) of a milking robot (150) to move to a desired position (pset) according to a desired velocity profile via: a feedforward module (110) producing predicted control signal(s) (cpred) based on a set vector (vset) specifying the desired velocity profile, a closed-loop controller (120), based on a modified position (Δp), producing primary control signal(s) (cprim) for controlling the end-effector (390) to the desired position (pset), and first and second summation modules (130; 140) deriving modified control signal(s) (cinput) to be fed to the milking robot (150) and deriving the modified position (Δp) respectively. The feedforward module (110) contains a trained artificial neural network with an input layer configured to obtain the set vector (vset), an output layer configured to provide the at least one predicted control signal (cpred), and a number of hidden layers interconnecting the input layer and the output layer. The respective nodes in said layers have weights that were assigned through a training process in which output signals (pout) from the robot (150) were used as training data and registered control signals for controlling the end-effector (390) of the milking robot (150) were used as reference data.