Page 24 - Read Online

Page 330 Lei et al. Intell Robot 2022;2(4):313­32 I http://dx.doi.org/10.20517/ir.2022.18

Table 2. Comparison of minimum path length, average path length, and standard deviation (STD) of path length with other models. The
values are reported for 30 executions
Test data set Model Min length (m) Average length(m) Length STD (m)
GA 9.47e+03 9.98e+03 4.46e+03
PSO 9.07e+03 9.25e+03 2.38e+03
Bays29 SOM 1.01E+04 1.28E+04 7.59E+03
ICA 1.09E+04 1.22E+04 9.87E+03
Proposed model 9.07e+03 9.07e+03 0
GA 2.39e+05 2.57e+05 1.15e+04
PSO 1.09e+05 1.17e+05 5.50e+03
KroA200 SOM 2.13E+05 2.60E+05 2.19E+04
ICA 2.12E+05 2.60E+05 6.67E+03
Proposed model 1.06e+05 1.06e+05 0
GA 1.37e+05 1.93e+05 6.30e+04
PSO 1.11e+05 1.14e+05 1.60e+03
PA561 SOM 1.01E+05 1.21E+05 1.84E+04
ICA 1.48E+05 1.51E+05 2.43E+03
Proposed model 1.03e+05 1.03e+05 0

12–14 s for the upper grid and 19–21 s for the lower grid. In a broiler barn, the cases shown in Figure 12A
took up the most proportion (420 grids), within which most contained no dead broilers, and the average robot
running time was 20–28 s in a single grid. Therefore, our proposed methods have a reasonable running time
in commercial-scale poultry barns, and they would be beneficial for rapidly detecting and collecting broiler
mortality. Simulation and comparison studies demonstrate the effectiveness and efficiency of the proposed
real-time robot safety-aware navigation.

5. CONCLUSION
A robotic system for detecting and collecting dead broilers in a barn is a promising direction for solving the
issueofbroilermortalityremoval. Aimingatthisvision, wedevelopedaninformativeplanningprotocol-based
multi-layer robot navigation system through detection and removal robots. The detection robot consists of
DCPP for constructing the overall trajectory; IPP for detailing the trajectory based on historical data, DCPP
direction, and obstacles; and a YOLO V4 dead bird detector for providing the precise locations of broiler
mortality along the trajectory. The removal robot receives the mortality location information and plans an
optimal ordered route by the HMTR scheme. The comparison and simulation results demonstrate the great
potential of the proposed methods for robot navigation, being useful tools for supporting precision broiler
management.

There are also many possible avenues for future work. A challenging extension is that we will integrate our
algorithmtotesttherobotsintherealsceneofthebroilerbarnanddevelopasuitablehuman–robotinteraction
platform for effective control. Another interesting topic is the application of multiple robots to cooperatively
search the environment to reduce the overall work time and simultaneously complete the functions of search,
identification, and removal to improve efficiency.

DECLARATIONS
Acknowledgments
The authors would like to thank the editor-in-chief, the associate editor, and the anonymous reviewers for their
valuable comments.

Authors’ contributions
Wrote and reviewed the manuscript: Lei T, Li G, Luo C, Zhang L, Liu L, Gates RS
All authors contributed equally.