A Control Classification of
Automated Guided Vehicle Systems
Automated guided vehicle systems (AGVS) are widely used for transporting material in manufacturing and warehousing applications. These systems offer many advantages over other forms of material transport. However, the design of these systems is complex due to the interrelated decisions that must be made and the large number of system design alternatives that are available. In particular, the design of the AGVS control system can be quite challenging, and it can dramatically affect the system cost and performance. This paper presents a classification of automated guided vehicle systems developed from a control perspective. The classification is useful for understanding the implications of the AGVS design decisions on the control system. It also provides the first step towards the development of a useful AGVS design aid that helps a system designer determine the most appropriate AGVS design for a particular application.
Significance: This paper deals with a classification scheme that provides a structured mechanism for organizing the relevant information about the design of the AGVS from a control perspective. It allows the system designer to determine how design decisions will impact the control complexity and provides the foundation for a design aid that helps in determining the most appropriate AGVS design for a specific application.
Automated guided vehicle systems (AGVS) are commonly used for transporting material within a manufacturing, warehousing, or distribution system. These systems provide for asynchronous movement of material through the system and are used in a wide variety of applications. They offer many advantages relative to other types of material handling systems, including reliable, automatic operation, flexibility to changes in the material handling requirements, improved positioning accuracy, reduced handling damage, easily expandable layout and system capacity, and automated interfaces with other systems.
The design of AGVS, however, can be very complicated because of the number of interrelated decisions must be made including determining the guide path layout and characteristics, the number and type of vehicles, the location, type, and buffer capacities of pickup/deposit stations, the operating procedures (e.g., vehicle dispatching and routing), the type of communications, and the type and characteristics of the control system (e.g., centralized, decentralized, zone or distributed). Most of these decisions also have an impact the design of the AGVS control system. The control system design is important since it greatly affects the system performance and overall installation and maintenance cost.
A classification scheme is needed to identify the relevant AGVS design alternatives from a control perspective. This classification shows the impact each of these decisions has on the controller design and also identifies the controller functionalities required for a particular system design.
The classification is useful to system designers in understanding the impacts the design decisions have on the control system so that the tradeoffs among the different design alternatives can be better evaluated.
In automated or semi-automated manufacturing systems, the AGVS controller is an integral part of the shop floor control system. The shop floor control system is responsible for routing products through the individual processing stations and interacting with the shop floor equipment and operators to affect production. The AGV’s role is to facilitate the transport of parts, tools, fixtures, etc., between individual processing centers as specified by the shop floor control system.
The AGV system is being made up of a supervisory controller (AGVSC) and subordinate vehicle controllers. Under this control paradigm, the vehicle controllers are responsible for the low-level drive system control (e.g., motors, transmission, etc.), and the supervisory controller is responsible for the higher-level system control functions including management of vehicle interactions. The control system functionality can be partitioned into planning, scheduling, and execution functions. The AGVSC is responsible for performing these functions. The overall structure of the AGVSC is shown in Figure 1.
In a hierarchical shop floor control system, the control functions have been partitioned into
· Planning: According to this architecture, planning is responsible for determining which tasks the control system should perform. This responsibility includes decomposing tasks into smaller sub-tasks and selecting the most appropriate task when alternatives exist. The AGVSC planning function is responsible for selecting an appropriate vehicle and determining the appropriate routing for that vehicle. Planning can be viewed as assigning tasks to individual vehicles, where the task identifies the path that the vehicle is to take. Planning is often referred to as routing and dispatching in the context of AGVS. Task assignment may be performed either dynamically, where the tasks are assigned to vehicles that are currently unassigned as requests for the service are received, or preplanned, where the tasks are assigned to vehicles without regard to their current assignment status . There are two broad categories of task assignment in the context of flexible manufacturing which demonstrate the effect of different assignment strategies on vehicle congestion and overall shop floor performance: work center-initiated, referring to vehicle selection from a set of competing idle vehicles; and vehicle-initiated, referring to the assignment of a work center to a vehicle from a set of competing work centers. Various assignment strategies have been evaluated for carrying a single load and multiple loads.
· Scheduling: The scheduling function is responsible for combining all of the individual vehicles’ routes into an overall sequence of vehicle segments or it can be said that, the planning function is responsible for breaking down the individual vehicle paths into smaller segments, and scheduling is responsible for sequencing the vehicles’ access to each segment. The scheduling function is also responsible for resolving vehicle conflicts or deadlocks and generating/updating expected start and finish times for the selected routes so a number of alternate routes for a given origin/destination pair may have to be evaluated before identifying a feasible route. Feasibility means that the selected route is not blocked and starting the vehicle along the route will not lead to irresolvable deadlock. Both blocking and deadlock are dynamic problems and can only be handled by considering other vehicles in the system.
· Execution: The execution function is responsible for interfacing with the subordinate vehicle controllers, initiating start-up and shut-down procedures, issuing commands for assigned activities, and monitoring for error detection and recovery. As such, the execution function provides the interface between the physical system and the software control system. This function depends only on the configuration of the physical system, whereas the planning and scheduling functions also depend on the production requirements.
Figure 1. A Detailed Schematic of AGVS Controller
Conflict in automated guided vehicle routing is said to occur when two or more vehicles are temporarily delayed if they are: (1) traveling along the same guide path but at different speeds, or (2) arrive at the same intersection from different guide path segments. The AGVSC must be able to resolve the conflict. There are several rules that can be followed for resolving conflict at intersections in unidirectional, single lane/aisle, guide path networks, such as: allowing departure of vehicles from an intersection on a first-come-first-served basis, restricting the first-come-first-served rule to vehicles transferring to the same path segment, prioritizing tasks and allowing departure based on these priorities, etc.
Deadlock resolution is another important support function of the AGVSC, which aids the scheduling function in evaluating feasible routes. Deadlock is a situation where further movement of a set of vehicles is inhibited due to the current status of the AGV system. There are three standard approaches to manage deadlock situations: prevention, avoidance, and detection and resolution. While the first two approaches ensure that deadlock will never occur, the last approach allows deadlock to occur and resolves it appropriately. It is the responsibility of the scheduling function to avoid or detect and resolve a deadlock in the AGVS to prevent eventual shop floor lockup.
There are two distinct deadlock situations that may arise in a general manufacturing system: (1) part routing deadlock; and (2) material handling deadlock. Part routing deadlock is a state when parts are assigned to various machines in such a way that any further progress of part movement is inhibited. Material handling deadlock is a state of deadlock when further movement of a material handling entity is inhibited due to the routing of the material handling entity. This situation is equivalent to a traffic gridlock in a city road network.
Consider the situation of an unmanned manufacturing system as depicted in the figure below which shows a manufacturing system with two machines (M1 and M2), an input/output buffer, and an AGV for material transfer between the machines and the input/output buffer. The figure depicts part 1 and part 2 currently being processed on machines M1 and M2, respectively, with the consequent destinations for part 1 and part 2 being machine M2 and machine M1, respectively. It is clearly known that any further movement is impossible without the intervention of an operator, and, hence, the system is in deadlock. Part routing deadlocks occur in unmanned systems with finite buffer capacities without proper planning and scheduling.
(a) Part Routing Deadlock (b) AGV Deadlock
Fig: Deadlock Situations in Manufacturing Systems
Material handling deadlock can occur in material handling systems such as AGVS or bridge cranes due to the conflicting routes of the material transporters. Considering the system shown in figure above with two vehicles on the same guide path and adequate buffer space at each machine for the parts on the vehicles the deadlock is due to the intrinsic operation of the automated guided vehicle system and not the part routing.
There are variety of solutions to the material handling deadlock problem which include design strategies such as building special spurs/sidings and operational strategies such as scheduling around the deadlock. Deadlock must be addressed either by preventing it at the design stage or building functionality into the controller to prevent or resolve it.
The purpose of the classification scheme described in this paper is to identify design alternatives of the AGVS that impact the controller design. This classification shows the impact each of these decisions has on the controller design. Based on the controller structure described above, the classification identifies the controller functionalities required for a particular system design.
The classification system has three basic levels as shown below.
1. Guide path determination
a) Static path
b) Dynamic path
2. Vehicle capacity
a) Single unit load
b) Multiple loads
3. Vehicle addressing mechanism
a) Direct address
b) Indirect address
Guide path Determination:
AGVS guide paths are determined in one of two ways:
Static determination: In a static guide path system, the vehicles use a set of predetermined paths between possible origins and destinations. The vehicles can use a variety of guidance mechanisms, such as floor embedded guide wires, chemical/optical sensor stripes, dead reckoning and position updating using targets or beacons based on a software map of the paths. A virtual path system, in which there is no physical guide path, and the supervisory controller determines the specific path from a set of fixed paths in a database, is considered a static guide path system. In this case, the controller function is the same, although the vehicle navigation system changes significantly.
Static guide path systems are further divided into
Unidirectional system: In a unidirectional system, vehicles are only allowed to travel in a single direction in a given lane. An aisle in the system can be divided into multiple lanes with each lane having its own direction of travel. Vehicle movement in one lane is independent of vehicle movement in another lane, e.g., the aisle is wide enough that two vehicles, one in each lane, can pass each other in the aisle. In this case, two lanes in an aisle may have opposing travel directions, but the system is still considered a unidirectional system, since each lane can be controlled independently. This restriction to unidirectional travel makes the system easier to control, since many of the deadlocking and collision avoidance problems are eliminated.
Bidirectional system: When short bidirectional segments are used at load/unload stations the vehicles enter and leave the segment from the same end and the segment is only large enough to allow one vehicle on it at any time. Therefore, the control system does not need the functionality of handling the bidirectional case. Also, vehicles are allowed to travel in both directions in the same lane. This functionality can be accomplished by providing turnaround points for vehicles or by using bidirectional vehicles, which are capable of moving forward and backward along the same guide path.
· There are obvious advantages to bidirectional systems in terms of increased efficiencies, productivity and a reduction in the number of vehicles required.
But the control of bidirectional systems is complex because of the contention of multiple vehicles for the shared guide path segments. In a bidirectional system, the controller must be able to manage the movement of vehicles to avoid or recover from deadlock situations.
Dynamic real-time determination: These systems use completely autonomous vehicles that are capable of determining a path using obstacle detection and avoidance systems. With dynamic paths, the vehicle is given the destination, which is a location that it knows about, perhaps specified relative to some world coordinate system. The vehicle then determines the path from its current position to the destination using its internal navigation scheme. These systems are not very prevalent in industrial applications and for the creation of autonomous vehicles and the dynamic planning of their motion there are many techniques which are being employed for navigation of the autonomous vehicles, including using vision systems, terrain topology estimation, and ultrasonic obstacle avoidance.
· Development of the sensory systems required for dynamic motion planning is also being researched. Since the vehicle maps the region and determines the path dynamically as it executes a task, the facility configuration can change and the AGVS control system can be updated with the new machines and the new locations relative to its coordinate system. The AGVS needs no knowledge of resources in the facility with which it does not pickup or drop-off loads. Therefore, these activities can change without affecting the AGVS control system.
AGV’s can be classified depending on the number of loads that the vehicle can simultaneously carry as either
· Single load: A “load” consists of a single “unit” carried by the vehicle from an origin to a destination. This unit may contain a number of distinct parts of the same or different types, e.g., assembly kits contained in a tote, but are considered a single load as long as all of the parts have the same origin and destination and the vehicle handles the tote as a unit. In a single load system, an idle, empty vehicle is selected for a task (i.e., assigned a load to deliver). The vehicle then travels from its current position to the pickup station to obtain the load and then travels to the destination station to drop off the load. Once the vehicle is assigned the task, it is not interrupted with another task assignment. The planning function must dispatch the vehicle and determine the route for the vehicle from its current location to the task origin and then to the task destination.
· Multiple load vehicles: For AGVS with multiple vehicle types, the system will be considered a multiple load system if any of the vehicles is multiple load vehicles. The distinction for the control system lies primarily in the planning function. In this system the assignment of tasks to vehicles is more complicated. Partially loaded vehicles may be interrupted in their current tasks to pickup additional loads. Assigning a vehicle to a task affects not only the load for this task but all of the other loads the vehicle may currently be carrying. Therefore, the planning and scheduling functions of the controller must determine the best vehicle assignment considering all of the loads and then replan and reschedule the vehicle’s movement to integrate the new tasks into the previously assigned tasks.
Vehicles with multiple load capability are widely available, particularly light load vehicles capable of carrying multiple totes, e.g., in electronics assembly application, for remote controlled automated load-haul-dump (LHD) vehicles in an underground mining operation. In this application, the vehicle is free to move to several digging locations to pick up dirt and other materials to be removed from the mine.
Vehicle Addressing Mechanism:
AGVS can be classified depending on the nature of the system operation as
1. Direct system: Direct address AGVS allow any vehicle to visit any station in the system. Each origin/destination pair of stations is directly served by the vehicles, much as in a city taxi service. In a direct address system, the planning function must route vehicles from their current location to their destination considering the current status of the system. Similarly, vehicles must be dispatched or assigned to tasks, since vehicles are not restricted to serving a subset of stations. These functions are complicated since the location of the vehicle is not known in advance but changes as the status of the system changes. This creates a dynamic planning problem in which the current state of the system must be taken into account in both the dispatching and route planning functions. The intuitive advantage of direct addressing is that individual loads can be transported more quickly using a direct route rather than an indirect route in which multiple vehicles could be involved.
2. Indirect address systems: In an indirect address system, each vehicle visits the load/unload stations in a fixed sequence, similar to a city bus service. In this case, the routes for each vehicle are determined in advance as part of the system design and are, therefore, not part of the controller planning function. In addition, dispatching in an indirect address system is straightforward. Since the vehicle visits the stations in a prescribed order, it picks up and drops off loads as it comes to the appropriate station. The only complication is if the controller can have the vehicle “wait” at a particular station for a load to arrive. Also, the route for each vehicle may not include every station in the system. That is, the stations are partitioned such that a vehicle serves some subset of the stations. This situation introduces a different type of planning problem, namely, how to “route” a load through the system.
It may occur that a load’s origin is served by one vehicle and a load’s destination is served by a different vehicle. In this case, the load must be transferred from one vehicle to another, much as people transfer between buses. Depending on the configuration of the system, the load may be handled by several vehicles before reaching its destination. In addition, there may be alternative “routes” for the load to take from its origin to its destination. The AGVS controller in an indirect address system must be able to plan this routing of loads through the system. The primary rationale for the single-vehicle loops is that the simplicity of the control makes the configuration attractive.
In this section, the AGVS control classification scheme described above is illustrated using examples taken from the published literature. Each AGVS design is categorized according to this scheme. The results are summarized in Table 1. In some cases, complete information is not available as a particular detail about the AGVS, which was perhaps unimportant to the published research, was omitted from the paper. Instances where a design class level was inferred from the paper are marked in the table.
Figure: AGVS Classification Scheme
AGV System Characteristics
(SU, SB, D)
(S or M)
(I or D)
Bartholdi and Platzman (1989)
Huang et al. (1989)
The classification scheme described above accommodates the wide variety of designs surveyed. It provides a structured mechanism for organizing the relevant information about the design of the AGVS from a control perspective. It allows the system designer to determine how design decisions will impact the control complexity. Obviously, the “best design” of an AGVS is system dependent.
The control system classification which is formulated in this paper provides helpful information to a designer.
This paper presents a classification scheme for automated guided vehicle systems which is developed from a system control perspective. The paper provides a discussion of the functionalities required of a generic AGVS controller. The classification scheme is then developed based on the impact the AGVS design alternatives have on the control system. The scheme is useful as a structured method for understanding the impact of design decisions on the control system. It also provides helpful information to the system designer about the impact of design decisions on the required controller functionality and resulting complexity. The ultimate goal is to use the classification scheme as a design aid. Further research is needed to develop a procedure based on this classification scheme that will help a user choose the most appropriate design, from among the many possibilities, based on the requirements and characteristics of their particular application. More importantly, it provides the foundation for the long-term development of an automated guided vehicle system design aid. Future research efforts will concentrate on building a methodology that uses the classification scheme to make design decisions that explicitly consider the tradeoff between controller complexity, and hence system cost, and system performance.
1. Bozer, Y.A., and Srinivasan, M.M., 1991, “Tandem Configurations for Automated Guided Vehicle Systems and the Analysis of Single-Vehicle Loops.”
2. Venkatesh, S., Smith, J.S., Curry, G., and Deuermeyer, B.L., 1994, “Deadlock Properties in Discrete Event Simulation,” Working Paper, Industrial Engineering Dept., Texas A&M University, College Station, TX.