Sensor Fusion-Based Cooperative Trail Following for Autonomous Multi-Robot System

. 2019 Feb 17;19(4):823. doi: 10.3390/s19040823

Algorithm 1 The feature fusion algorithm using SVM/SOFTMAX/four-layer DNNs.

Input: Image

X_{g r o u n d}

acquired by the ground robot, image

X_{a e r i a l}

obtained by the aerial robot, the trained ground and aerial classifiers to extract the features from the vision input, and the trained feature fusion modules SVM/SOFTMAX/four-layer DNNs to complete the feature fusion process.

Output: Class label C defined in the problem formulation, which is then transferred as the flight command of the robot.

1:
Initialize and take off from the ground and aerial drones (robots).
2:
The ground robot gets the vision input.
3:
$C_{g r o u n d}$ = $C F_{g r o u n d} (X_{g r o u n d})$
4:
The ground robot gets the list $(P G_{s t r a i g h t}, P G_{l e f t}, P G_{r i g h t})$ from the SOFTMAX layer of the trained ground classifier.
5:
$ψ = m a x (P G_{s t r a i g h t}, P G_{l e f t}, P G_{r i g h t})$
6:
if $ψ < T h r e s h o l d$ then
7:
The ground robot extracts the low-altitude features $f_{g r o u n d}$ using the trained ground classifier from the low-altitude vision input and sends a classification request signal to the aerial robot.
8:
The aerial robot extracts the high-altitude feature $f_{a e r i a l}$ using the trained aerial classifier from the high-altitude vision input and sends the features to the ground robot.
9:
The ground robot sends the low-altitude and the high-altitude features to the SVM/SOFTMAX/four-layer DNNs modules to implement the feature fusion process.
10:
The ground robot gets the final decision based on the integrated representation from the feature fusion modules. $C = C_{S V M / S O F T M A X / 4 - l a y e r D N N s}$
11:
else
12:
$C = C_{g r o u n d}$
13:
end if
14:
Both the aerial robot and the ground robot take the C command.