Performance Assessment of PRIDE in Manufacturing Environments

This paper describes PRIDE (Prediction in Dynamic Environments), a multi-resolution and hierarchical framework. PRIDE was developed as a test bed to assess the performance of autonomous vehicles in the presence of moving objects in a simulated environment. By simulating scenarios in which moving objects are prevalent, a designer of an autonomous vehicle can test the performance of their path planning and collision avoidance algorithms without having to immerse the vehicle in the physical world. This framework supports the prediction of the future location of moving objects at various levels of resolution, thus providing prediction information at the frequency and level of abstraction necessary for planners at different levels within the hierarchy. Previous works have demonstrated the reliability of PRIDE to simulate on-road traffic situations with multiple vehicles. To provide realistic scenarios, PRIDE integrates a level of situation awareness of how other vehicles in the environment are expected to behave considering the situation in which the vehicles find themselves in. In recent efforts, the PRIDE framework has been extended to consider production logistics in dynamic manufacturing environment while focusing on the scheduling of material transportation system. To demonstrate the characteristics of the PRIDE framework, this paper illustrates real-time navigation of Automated Guided Vehicles (AGVs) at different locations in a dynamic manufacturing environment. Moreover, using the high-fidelity physics-based framework for the Unified System for Automation and Robot Simulation (USARSim), this paper analyzes the performance of the PRIDE framework on a set of realistic scenarios. 1. INTRODUCTION From traditional and well-established applications in the automotive industry to emerging applications such as material handling, palletizing, and logistics in warehouses, the use of mobile robots can increase productivity whilst ensuring personnel safety. Automated Guided Vehicles (AGVs) represent an integral component of today’s manufacturing processes. They are widely used on factory floors for intra-factory transport of goods between conveyors/assembly sections, parts/frame movements, and truck-trailer loading/unloading. According to Bishop Consulting’s report [1] on AGV Industry Next-Generation Technology Priorities, “In the eyes of the system vendors, the most prominent technology development area is in moving from today’s AGVs, which require highly structured environments and reference markers installed throughout the plant, to operating in less structured or unstructured environments. In fact, the site preparation required to install these reference markers is a significant portion of the system cost ...”. To offset prohibitively expensive maintenance and installation costs, and thus expand the AGV’s markets and utility beyond what is possible today, it is evident that the dependency on infrastructure is to be minimized (if not eliminated). To achieve this goal and to be able to cope with unstructured, dynamic environments, predicting future positions of moving objects in factory environments are critical enablers for widespread use of AGVs. The research interest of this paper deals with path planning of AGVs using the PRIDE (PRediction In Dynamic Environments) framework. PRIDE is a multi-resolution, hierarchical framework that provides an autonomous vehicle planning system with information required to perform path planning in the presence of moving objects. PRIDE incorporates multiple prediction algorithms into a single, unifying framework. To date, we have applied this framework to simulate the prediction of the future location of autonomous vehicles during on-road driving. The PRIDE algorithms are not limited to on-road driving and can be ported to other domains. As such, this paper illustrates how PRIDE has been extended to path planning of AGVs in manufacturing environments. In the factory environments used by PRIDE, the autonomous vehicles are industrial robots (e.g., unit loaders, forklifts) performing collision-free intra-factory activities, including transport of goods between conveyors and assembly sections, parts and frame movements, and trucktrailer loading/unloading. The remainder of this paper is organized as follows: Section 2 gives an overview of the PRIDE framework. Section 3 describes the different features added and modified in PRIDE to port this framework to manufacturing. Section 4 details a manufacturing moving object ontology (M2O2), an ontology used to plan the paths of the other AGVs. Section 5 discusses the performance of the PRIDE framework through three scenarios involving two AGVs and Section 6 concludes this paper and gives an overview of the future work. 2. THE PRIDE FRAMEWORK PRIDE is a multi-resolution hierarchical framework that provides an autonomous vehicle planning system with information required to perform path planning in the presence of moving objects. This framework supports the prediction of the future location of moving objects at various levels of resolution. PRIDE is based on the 4D/RCS architecture [2], which provides a reference model for unmanned vehicles on how their software components should be identified and organized. The PRIDE framework provides moving object predictions to planners running at any level of the 4D/RCS hierarchy at an appropriate scale and resolution. The underlying concept of PRIDE lies in the incorporation of multiple prediction algorithms into a single, unifying framework. At the higher levels of the framework, the prediction of moving objects needs to occur at a much lower frequency and a greater level of inaccuracy is tolerable. At these levels, moving objects are identified as far as the sensors can detect and a long-term (LT) prediction algorithm predicts where those objects will be at various time steps into the future. Higher-level reasoning processes need a global representation of the environment to compute the future location of an AV. PRIDE uses the road network database (RND) [3] to access different information about the road networks, including individual lanes, lane markings, intersections, legal intersection traversability, etc. The lower levels of the framework use estimation theoretic short-term (ST) predictions based on an Extended Kalman Filter (EKF) to predict the future location of moving objects with an associated confidence measure. Complete details on the LT and ST prediction algorithms can be found in previous efforts [4]. PRIDE currently integrates the Mobility Open Architecture Simulation and Tools (MOAST) framework along with the Unified System for Automation and Robot Simulation (USARSim) [5]. This integration provides predictions incorporating the physics, kinematics and dynamics of AVs involved in traffic scenarios. MOAST is a framework that provides a baseline infrastructure for the development, testing, and analysis of autonomous systems1. MOAST implements a hierarchical control technique, which decomposes the control problem into a hierarchy of controllers with each echelon (or level) of control adding additional capabilities to the system. USARSim is a high-fidelity physics-based simulation system that provides the embodiment and environment for the development and testing of autonomous systems. USARSim utilizes high-quality 3D rendering facilities to create a realistic simulation environment that provides the embodiment of a robotic system. The system architecture on the integration of PRIDE with the MOAST and USARSim frameworks is described in previous work [6]. PRIDE also handles drivers' aggressivity. In this context, the aggressivity represents the style and driving preferences of a driver. For example, one would likely assume that a conservative driver will remain in his lane whenever possible and will keep a gap between his vehicle and the leading vehicle. Conversely, an aggressive driver would have a higher probability of changing lanes and would be more apt to tailgate the leading vehicle. One may also find that the aggressivity of the driver may change over time, e.g., the driver can be very aggressive when trying to get to a certain lane, but become more passive when he gets there. The PRIDE framework addresses all the driver types and situations mentioned above. Experiments and corresponding results performed on aggressivity can be found in previous work [7]. 3. FROM ON-ROAD TO MANUFACTURING ENVIRONMENTS The PRIDE algorithms are not limited to on-road driving and can be ported to other domains. In this paper, the PRIDE framework deals with path planning in manufacturing environments. In a factory setting, the autonomous vehicles are AGVs (e.g., unit loaders, forklifts) performing collision-free intra-factory activities, including transport of goods between conveyors and assembly sections, parts and frame movements, and truck-trailer loading/unloading. To handle these systems, some features of the current framework have to be modified. The RND structures must comply with factory settings and safety. New maps are built to accommodate industrial robots requirements. For instance, hazardous areas are specified and loading/unloading stations are setup for goods deliveries around the factory. 























































 1
Autonomous systems in this context refer to embodied intelligent systems that can operate fairly independently from human supervision.
 To be able to handle new missions in manufacturing environments, the long-term prediction algorithm is modified as well. During on-road driving, the autonomous vehicles were not asked to reach a specific target. Instead, the PRIDE algorithms focused on the actions the autonomous vehicles would take based on the situations they find themselves in. In manufacturing settings, AGVs have to reach different targets to load and unload supplies for delivery. For better productivity, AGVs have to move along collision-free paths with the minimum time. To reduce the time during missions, PRIDE uses the algorithm of Dijkstra [8] to compute the shortest path from the start position to the target position for each AGV. Figure 1 depicts the process overflow of the PRIDE prediction algorithm. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 As illustrated in Figure 1, the PRIDE algorithms check if the current vehicle has to reach a target (1), which is specified in a xml file. If a target is not specified, the vehicle chooses random paths (2) and the PRIDE algorithms send waypoints to control the AGV (7). If the AGV has to reach a target, the Dijkstra’s algorithm is used to compute the shortest path from the current position of the AGV to the Figure
1.
Process
overflow
of
the
long­term
prediction
 algorithm.
 Control
the
AGV
 Remove lane of interest Vehicle of interest Another vehicle in the way Compute the shortest path Use
random
 paths
 Reaching a target