Page 8 - Read Online

Page 314 Lei et al. Intell Robot 2022;2(4):313­32 I http://dx.doi.org/10.20517/ir.2022.18

ficiently and safely, thus being a useful component for robotics. The effectiveness and robustness of the proposed
methods are validated through simulation and comparison studies.

Keywords: Directed coverage path planning, informative planning protocol (IPP), broiler mortality, YOLO V4, hub-
based multi-target routing (HMTR) scheme

1. INTRODUCTION
1.1. Background
Global broiler production has been growing since the 1960s; in the United States alone, over 9.22 billion broil-
[1]
ers were produced in 2020 at a value of 21.7 billion dollars . A modern broiler barn holds 25,000 to 100,000
broilers on the open litter floor. Within the intensive production system, broiler mortality could be largely
[2]
due to disease and metabolic problems and unsuitable environmental conditions and management . The
mortality rate is commonly 5% in an eight-week production cycle [1] and can reach 10% without appropriate
[3]
management . As a daily task, producers need to collect and remove broiler mortality timely, which is an
arduous, time-consuming, and unpleasant task. A farmer spends hours daily identifying, gathering, and trans-
porting dead birds in a typical broiler barn (e.g., 25,000 birds/barn). Manual mortality collection is relatively
simple when broilers are young because they are lightweight and small. However, older broilers are bigger and
weigh more, and, to avoid overwork, farmers may deposit the mortality in scattered piles within the broiler
[3]
barn for subsequent removal . A dead bird residing on a litter floor can result in higher levels of bacterial
[3]
accumulation and increase the risk of disease spread via direct contact or vector transmission . Additionally,
caretakers risk their health if exposed to harsh working environments (e.g., concentrated ammonia and dust)
[2]
for a long term . Therefore, autonomous robotic systems for collecting broiler mortality are warranted to
improve labor efficiency and reduce biosecurity issues.

Robotic systems have been developed to facilitate poultry production [4,5] . Vroegindeweij et al. developed a
[6]
PoultryBot to avoid obstacles and pick up floor eggs in cage-free hen housing systems . Li et al. developed an
[7]
egg picking robot based on a deep learning detector and a robot arm . Some poultry robots are commercial-
ized to help farmers monitor environments, inspect bird health and welfare status, stimulate bird movement,
aerate litter, and disinfect poultry barns. However, robots for poultry mortality removal are still not commer-
[8]
cially available. Many aspects need to be considered for developing such a robot . Among them, robot path
planning is one of the most essential parts of effective and efficient determination of robot routes [9,10] .

1.2. Related work
Many approaches have been proposed to achieve reliable autonomous robot motion planning, such as ant
colony optimization (ACO) [11–13] , fireworks algorithm (FWA) [14] , bat-pigeon algorithm (BPA) [15] , graph-
based method [16] , and neural network models [17–19] . Lei et al. proposed a hybrid model to optimize the
trajectory of the global path using a graph-based search algorithm associated with an ant colony optimiza-
tion (ACO) method [11] . A hybrid fireworks algorithm with LIDAR-based local navigation capable of gen-
erating short collision-free trajectories in unstructured environments was developed [14] . The cuckoo search
algorithm has also been successfully applied to the efficient and safe navigation of robots [20] . A bat-pigeon al-
gorithm [15] was developed with crackdetection-driven autonomous vehicle navigation andmapping, in which
a local search-based bat algorithm and a global search-based pigeon-inspired optimization algorithm are ef-
fectively integrated to improve the speed and performance of robot path planning and mapping. Wang and
Meng [16] suggested a nonuniform sampling technique, which efficiently computes high-quality collision-free
paths based on a generalized Voronoi graph. A biologically motivated neural network model using a shunting
equation was proposed by Yang and Luo [17] for real-time path planning with obstacle avoidance. Luo et al.
extended the model of trajectory planning with safety consideration in conjunction with the virtual obstacle