Page 10 - Read Online

Page 316 Lei et al. Intell Robot 2022;2(4):313­32 I http://dx.doi.org/10.20517/ir.2022.18

method for the localization robot could pinpoint targets and avoid unnecessary searching trajectories, saveing
the robot time and energy.

After the locations of dead broilers are precisely and efficiently identified, the removal robot is deployed to
collect the mortalities. To save time and energy, the total length of the path to remove multiple dead broilers
in succession should be minimized [31] . In that regard, a new hub-based multi-target routing (HMTR) scheme
is developed in this paper, which originates from the row-based environment routing of broiler barn. By intro-
ducing hub grids, computational efficiency is improved for distance calculations between targets. Based on the
locations of the dead birds, the best visiting sequence is determined while planning collision-free trajectories
simultaneously. Furthermore, a reactive local navigator is developed to dynamically update the path and map
in real time to avoid moving obstacles and unforeseen obstacles in the dynamic environment. This is of great
help to robot safety and obstacle avoidance and prolongs the life of the robot.

Overall,weproposeasystemfordetectingandremovingbroilermortalitywithtworobots. Thedetectionrobot
consists of an informative directed coverage path planner and a You Only Look Once (YOLO) V4-based dead
bird detector. The removal robot collects dead broilers by the hub-based multi-target path routing (HMTR)
scheme. Meanwhile, with the reactive local navigator, the robots avoid obstacles and reach the target position
through the real-time updated information of the onboard LIDAR sensors.

The major contributions of this paper are summarized as follows:

• A multi-layer robot navigation system for search, locate, and remove (SLR) operations is developed for the
removal of dead broilers in a poultry barn to reduce the daily laborious and time-consuming work.
• A directed coverage path planning (DCPP) method integrated with an informative planning protocol (IPP)
is developed to efficiently search targets in a large-scale workspace, which is initially decomposed into grids,
and the optimal coverage directions are obtained based on the workspace. The detection robot adaptively
explores the workspace to generate the trajectory in the grids in light of the historical data of searched
targets and the coverage directions.
• The informative planning protocol (IPP) is developed to integrate coverage directions devoted to rapidly
achieving spatial coverage with the least estimation uncertainty in the decomposed grids.
• To efficiently reach the targets, a new hub-based multi-target path routing (HMTR) scheme is proposed to
row-based environments. The total visiting distance is minimized through obtained targets to generate an
optimal sequence to connect all the targets.
• Practically, to apply this navigation system in the poultry barn, multiple robots with this developed naviga-
tion system are utilized, consisting of a dead broiler search robot to search and locate the dead broilers and
a dead broiler removal robot to collect the dead broiler afterwards.

2. PRELIMINARY DEFINITION
2.1. Workspace setup and robot configuration
The complex barn environment is simplified into a rectangular robot workspace (152 m long × 12 m wide)
based on typical United States broiler barn dimensions [32] . The workspace where the robot needs to collect
dead broilers consists of multiple feeding and drinking lines that run the length of the barn, start and end
positions of the robot, and dead and live broilers that are randomly located in the barn. Two feeding and three
drinking lines are arranged in parallel in the middle, each measuring 140 m long and 0.3 m wide [Figure 1].
The distance between lines is 1.75 m, and endpoints of the lines are 6 m away from the ends of the barn. The
start and end positions are located near the gate of the barn and could be altered in a real-world application.
Live/dead birds are simplified as 20 cm diameter circles. With a stocking density of nearly 929 cm /bird (1
2
ft /bird), the barn is assumed to have 20,000 birds. The workspace is illustrated in Figure 1.
2