Robot Machine Learning

Many robot predictive maintenance applications require being able to decide whether a new observation belongs to the same distribution as existing observations or should be considered as an outlier.
More information may be found here: https://scikit-learn.org/stable/modules/outlier_detection.html

Create Robot Models

We will create one "nominal" robot and one "defective" robot.
The defective robot will include a small offset at the TCP (e.g., imagine the tool is wearing out).



In [1]:

    
from pybotics.predefined_models import ur10
from pybotics.robot import Robot

nominal_robot = Robot.from_parameters(ur10())

defective_robot = Robot.from_parameters(ur10())
defective_robot.tool.position = [0.1, 0, 0]

Create Database

We need to "collect" a large amount of data for machine learning applications



In [19]:

    
import numpy as np
from pybotics.geometry import matrix_2_vector

joints = np.random.uniform(low=-np.pi, high=np.pi, size=(1000, 6))
nominal_poses = np.array(
    [matrix_2_vector(nominal_robot.fk(j)) for j in joints])
defective_poses = np.array(
    [matrix_2_vector(defective_robot.fk(j)) for j in joints])

inliers = np.hstack((joints, nominal_poses))
outliers = np.hstack((joints, defective_poses))
outliers = outliers[:20]

data = np.vstack((inliers, outliers))

truth = np.ones(len(data))
truth[-len(outliers):] = -1

Outlier Detection

The data contains outliers which are defined as observations that are far from the others (i.e., defective TCP).
Outlier detection estimators try to fit the regions where the training data is the most concentrated, ignoring the deviant observations.



In [22]:

    
from sklearn.neighbors import LocalOutlierFactor

# create classifier
clf = LocalOutlierFactor(contamination='auto')

# predict
y_pred = clf.fit_predict(data)

# compute accuracy
n_errors = (y_pred != truth).sum()
accuracy = (1 - n_errors / len(y_pred)) * 100
print(f"Accuracy: {accuracy:.2f}%")









    



Accuracy: 98.04%