IEEE Robotics & Automation Magazine - March 2018 - 56

(a)

(b)
Figure 1. (a) and (b) AGVs sharing the environment with human
operators.

Figure 2. AGVs moving goods in a modern factory warehouse
with high production volumes.

In this article, we present the main technological developments achieved during the Plug and Navigate (PAN)Robots project (http://www.pan-robots.eu), which aimed
at increasing the autonomy and efficiency of AGVs used
for industrial logistics in environments shared with
human operators. The main contribution of this article is
to provide a system-level overview of those achievements,
demonstrating how they can contribute in increasing the
applicability of AGV systems.
Scenario and Related Works
In this article, we consider AGV systems used for transporting pallets of goods in automated warehouses, as
shown in Figure 2. Specifically, we focus on common
manufacturing plants characterized by large production
batches, such as beverage companies. Typically, in these
environments, a few tens of vehicles are utilized for goods
transportation (from the production machines to the
warehouse, within the warehouse itself, or to the shipment
56

IEEE ROBOTICS & AUTOMATION MAGAZINE

March 2018

area). While traditional installations consist of manually
driven forklifts, advanced solutions based on AGVs are
becoming increasingly popular [4].
In these applications, AGVs share the environment with
other entities, such as pedestrians (i.e., human operators)
and manually driven vehicles (e.g., forklifts) [5], [6]. As a
consequence, safety concerns are of paramount importance; the environment is populated with dynamic entities,
and avoidance of collision needs to be guaranteed. A typical
solution for achieving this objective is in the use of safety
laser scanners [7], which allow each AGV to detect the
presence of obstacles in its vicinity and opportunely stop to
avoid collisions or replan the path to be traveled [8]. As discussed in [9], safety laser scanners do not allow AGVs to
classify detected obstacles and, subsequently, to make highlevel decisions, such as stopping in the presence of a human
(whose behavior is unpredictable) and circumventing a box
(which does not move).
Computer-vision-based techniques for dynamic obstacle
detection have been extensively studied in the last few years,
particularly in the context of autonomous vehicles [10], [11].
Because the environment is populated by human operators,
illumination is always needed. Hence, techniques based on
computer vision can be effectively utilized.
Industrial environments are typically very congested; as
shown in Figures 1 and 2, AGVs move through corridors
and racks to collect and deliver pallets of goods. To effectively plan the motion of the AGVs and react to unpredictable
events (e.g., the presence of a pedestrian) in the correct
manner, multiple viewpoints as well as a global view of the
environment are necessary. A sensing system was introduced in the PAN-Robots project that exploits a composition of onboard and offboard sensing systems to acquire
data from the environment that are then gathered in a centralized data fusion system (see the "Advanced Sensing System" section).
The motion of the AGVs then needs to be coordinated
through the environment in such a way that the requested missions (i.e., transportation of a pallet of goods from
one location to another) are fulfilled. Generally, two different philosophies can be followed to coordinate the
motion of the AGVs: decentralized and centralized. In
decentralized coordination strategies, each vehicle defines
its own path independently, based on locally available
information. Coordination among vehicles is then handled locally. While those strategies are known to scale well
for large-scale fleets [12], [13], they typically do not provide a complete solution. Hence, efficiency cannot always
be guaranteed. Since, in industrial applications, the overall
efficiency is paramount [4], we consider a centralized
coordination strategy that also incorporates information
acquired by the centralized data fusion system. As is
typically done in industrial applications, the proposed
coordination strategy considers the AGVs constrained
to move along a road map [14], which is a set of (virtual)
paths that let the AGVs reach any location of interest in

http://www.pan-robots.eu

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2018