Existing traffic signal control methods primarily rely on data from a single sensor modality, limiting their ability to comprehensively capture the spatiotemporal features of the traffic network. This leads to a mismatch between traffic signal timing plans and dynamic traffic demands, resulting in underutilization of road capacity. To address these limitations, this paper proposes a traffic signal-vehicle cooperative control method (SVCM) based on multimodal heterogeneous subgraph learning. Specifically, we first design a hybrid-strategy action space with dual-state transitions, enabling the agent to dynamically switch between traffic signal control and cooperative traffic signal-vehicle control schemes according to real-time traffic conditions. Then, by analyzing vehicle steering requirements, driving efficiency factors, and vehicle heterogeneity, we construct a multi-factor weighted decision-making model for vehicle lane-changing. It adjusts the lane-changing probability to balance traffic density among lanes. In addition, to enhance the perception and utilization of the spatiotemporal features of the road network, we designed a multimodal heterogeneous subgraph attention network to integrate multi-source traffic data. This approach more accurately captures key regional features and provides a foundation for state information perception in the aforementioned collaborative control strategies. Experimental results demonstrate that the proposed method, SVCM, outperforms state-of-the-art approaches in both simulated and real-world traffic scenarios. Compared with the best baseline method, SVCM reduces the average queue length by 24.38% to 47.14% and decreases average lane occupancy by 21.09% to 39.72%. This work provides new insights into traffic signal-vehicle cooperative control by integrating multimodal feature information.

This research investigates the control difficulties related to heterogeneous traffic flow in container terminals, featuring both connected and automated vehicles (CAVs) and human-driven vehicles (HDVs). The lack of signal control at intersections and the unpredictable routes taken by HDVs make the efficient transport of containers within a terminal quite challenging. To tackle this issue, we present an integrated traffic control policy aimed at enhancing transportation efficiency. For each unsignalized intersection, a virtual token ring system is introduced to manage the passage of vehicles, using a back pressure-based algorithm and specific token delivery rules to determine phase sequence and duration. Furthermore, we introduce an improved back pressure-based dynamic routing method for CAVs, which allows for the selection of roads with shorter travel time while adhering to a travel distance constraint when crossing an intersection. This approach aims to minimize disruptions from HDVs, reduce travel time, and prevent excessively long travel distances. Multiple experiments are conducted to verify the proposed method’s effectiveness.

Multiple mobile robot systems (MMRSs) play an important role in workshops, storage and other scenarios. A complex MMRS can contain heterogeneous robots, and even the robot sizes are not fixed, making it more difficult to avoid collisions. As a widely studied method, the zone-controlled method is not accurate enough for a system with heterogeneous and variable-sized robots, and it increases the difficulty of applying MMRS. Based on the concept of glued nodes, this paper proposes a more accurate calculation method than the zone-controlled method to avoid collisions among heterogeneous and variable-sized robots. In order to improve the efficiency of calculating glued nodes, this paper proposes a time-to-space method, which has a constant complexity after the system runs long enough. In addition to collision avoidance, this paper also studies the problem of giving way of idle robots. This paper proposes a method which can detect those idle robots blocking the working robots and plan paths for the idle robots to find avoidance nodes to give way to the working robots. Simulation experiments are carried out based on a real automated production line scenario, and the experimental results prove the effectiveness and efficiency of the proposed methods.Note to Practitioners—In a complex MMRS, the size of different robots can be different, and even the size of a robot is different when loaded and unloaded. One of the motivations of this paper is to avoid collisions in such a complex MMRS. A collision avoidance method is proposed to avoid collisions among robots by allocating nodes in real time. At the same time, this paper also addresses the problem of idle robots blocking working robots. If an idle robot blocks some working robots, an avoidance node will be found for the idle robot to give way to the working robots. The methods proposed in this paper can be applied to the roadmap of any structure. This means that in a scenario, regardless of the structure of roadmap and the different sizes of robots, the methods can be applied directly. The methods have been applied to many practical industrial projects in warehousing, manufacturing and other scenarios, and greatly reduces the difficulty of applying MMRS. In future research, there is a need to improve the efficiency of the methods so that they can be applied to a large-scale system.

Multi-mobile robot systems (MMRSs) are widely used for transportation in industrial scenes such as manufacturing and warehousing. In an MMRS, motion coordination is important as collisions and deadlocks may lead to losses or system stagnation. However, in some scenarios, robot sizes are different when loaded and unloaded, which means that the robots are variable-sized, making motion coordination more difficult. The methods based on zone control need to first divide the environment into disjoint zones, and then allocate the zones statically or dynamically for motion coordination. The zone-control-based methods are not accurate enough for variable-sized multi-mobile robots and reduce the efficiency of the system. This paper describes a motion coordination method based on glued nodes, which can dynamically avoid collisions and deadlocks according to the roadmap structure and the real-time paths of robots. Dynamic features make this method directly applicable to various scenarios, instead of dividing a roadmap into disjoint zones. The proposed method has been applied to many industrial projects, and this study is based on some manufacturing projects for experiments. Theoretical analysis and experimental results show that the proposed algorithm is effective and efficient.

In recent years, automated guided vehicles (AGV) are widely used to sort and transport parcels in logistics warehouses. The deployment of AGVs can improve storage efficiency and free human labour greatly. However, as the number of AGVs grows, the computational complexity and deadlock occurrence rate increase simultaneously, making it extremely difficult to coordinate AGVs’ movements in real time. In this paper, we first present a hierarchical motion coordination system based on event-triggered colored elementary net (ETCEN). The primary aim of the system is to coordinate AGVs in real time regardless of the system’s scale. Then, we describe deadlocks by the ETCEN model and classify them into two categories - active deadlocks and passive deadlocks. Active deadlocks are prevented dynamically by controlling the movements of AGVs, while passive deadlocks are resolved by an improved path planning strategy. The entire system relies on a two-layer architecture, thereby improving flexibility and scalability. The proposed algorithms are validated by simulations and applications. Experiment results demonstrated that our approach can coordinate multi-AGV systems, avoid collisions and prevent deadlocks effectively. Note to Practitioners—This paper was motivated by the problem of coordinating multi-AGV system in warehouses, especially the tricky problem of preventing deadlocks in large-scale applications. Existing approaches mainly focused on collision-avoiding strategies, while paying less attention to the deadlock-preventing problem. As a result, the feasibility of current deadlock prevention methods strongly depends on the topology of the environment, and the computation time will surge exponentially as the scale of AGVs grows. This paper suggests a general approach towards solving two types of deadlocks (active deadlocks and passive deadlocks) based on Petri net theory. In this paper, a multi-AGV coordination model is presented and described thoroughly. Then we use the model to prove our theories and introduce our collision-avoiding and deadlock-preventing strategies. A two-layer architecture is designed to support the expansion of AGV fleet, making our system highly scalable. Both experiments and applications suggest that this approach is feasible and effective. In future research, we will focus on further optimizing the coordination and routing algorithm to improve the system throughput.

This article considers intelligent fuel cell/battery hybrid vehicles (FCHVs) that can make autonomous decisions at both the vehicle and powertrain levels. Since the vehicle and powertrain level dynamics are inherently integrated, we propose an integrated motion and powertrain model predictive control approach for intelligent FCHVs by jointly optimizing the vehicle acceleration and fuel cell current. The control goals are to achieve vehicle mobility, minimal hydrogen consumption, and battery state-of-charge maintenance within system constraints. The main challenge in an integrated control is that the electric motor can operate in both propelling and generating modes coupling with vehicle and powertrain states. This hybrid operation is handled by the mixed logical dynamical modeling resulting in a mixed integer nonlinear control problem. To relieve the possible heavy computational burden, two simplification approaches are proposed: hierarchical control and successive linearizations. Two standard driving cycles and a typical vehicle cruising scenario are employed to test the effectiveness of the proposed modeling and control algorithms. Simulation results show that the hierarchical linear control is more suitable for real-time applications with comparable control performance with that of the integrated control. However, additional constraints must be carefully designed to compensate for the ignored coupling dynamics and constraints.

Automated guided vehicles (AGVs) are being extensively used for transportation and distribution of materials due to their high-efficiency. However, the vehicle-collision free problem is challenging since, when modeling these systems, there are indistinguishable and uncontrollable events due to the limited sensors and actuators. This paper proposes an approach to the design of a maximally permissive (optimal) controller to prevent vehicles from any collision based on Petri nets (PNs). For a typical class of AGV systems, a system modeling algorithm is presented using labeled PN, where indistinguishable events are represented by a set of transitions carrying the same label, and an uncontrollable event by an uncontrollable transition. By virtue of the PN model, the collision-free problem is formalized as a conjunction of linear constraints that are converted into admissible ones by an algorithm such that the computational overhead due to uncontrollable events is significantly reduced. In turn, a method is developed to compute the set of consistent markings for an observed sequence of labels that represent signals generated by sensors. Finally, given an observed sequence, a maximally permissive control action is computed to enforce a conjunction of admissible linear constraints based on the set of consistent markings. The approach well addresses the challenging issues caused by indistinguishable and uncontrollable events. A typical AGV system is utilized to illustrate and verify the theoretical results throughout the work.

It has been an active research area to develop robust supervisory control policies for production systems with unreliable resources. So far, most methods for robust deadlock resolution apply only to systems without flexible routes, where each processing step of any part type requires a unique prespecified resource. In this paper, we address deadlock avoidance control problem in production systems with both failure-prone resources and flexible routings, which allow that a part has options when deciding the resource acquisition at each step. This paper presents properties that a controller with robustness must satisfy. Specifically, at any system reachable state, neither the failed resources nor part instances trapped in these resources should have too much detrimental effect on the other portions of the system. Thus, the full range of part types' production could be assured at all time. After defining the notions of reduced system and reduced state with respect to unreliable resources, we identify and prove conditions for determining whether or not the state resulting from the occurrence of an event is feasible. Subsequently, we develop a method for robust deadlock avoidance, which uses the solutions to state safety checking problem for the reduced production system with only reliable resources. An illustrative example shows the effectiveness of this method. Finally, we conduct a comparison investigation of some representative approaches in the literature about robust supervisory control for deadlock resolution in resource allocation systems with routing flexibility.

Deep learning has achieved impressive success in natural language processing. However, most previous models are learned on the specific single tasks, suffering from insufficient training set. Multi-task deep learning can solve this dilemma by sharing the part of parameters, improving generalization. The common multi-task deep learning model consists of the shared layer and the task specific layer. In this paper, we attempt to enhance the performance of shared layer and proposed two variants based on convolutional neural networks. The first model is Agent Model-Direct Concatenate, where each task is assigned with a separate convolutional neural network for extracting the common and task specific features simultaneously. The second model is Agent Model-Gating Concatenation, where the task specific layer could automatically decide the information flow of each element of the output of shared layer. The two networks are trained jointly over three pair-wise groups of movie review data sets. Experiments show the effectiveness of our two networks, inspiring a potential direction for the related research of multi-task deep learning.

In this paper, we propose a hierarchical model predictive control (MPC) approach for signal split optimization in large-scale urban road networks. To reduce the computational complexity, a large-scale urban road network is first divided into several subnetworks using a network decomposition method. Second, the MPC optimization problem of the large-scale urban road network is presented, in which the interactions between neighboring subnetworks are described with interconnecting constraints. To coordinate the subnetworks, Lagrange multipliers are introduced to deal with interconnecting constraints among subnetworks, and an augmented Lagrange function is constructed. Then, based on dual optimization theory and a decomposition strategy, the dual optimization problem of the original MPC problem is divided into several new subproblems. In addition, we develop a coordination algorithm based on an interaction prediction approach to coordinate the resulted subproblems with a two-level hierarchical structure. Finally, experimental results by means of simulation on a benchmark road network are presented, which illustrate the performance of the proposed approach.

Arterial signal coordination based on a bandwidth criterion is one of the most effective and important signal coordination strategies for urban traffic signal control. In this paper, we propose a coordination methodology for arterial traffic signal control based on a novel two-way bandwidth maximization model. One of the important features of our method is that the queueing process has been considered. Different from most existing models in which the queue clearing time was neglected or artificially set by experienced engineers, the queue clearing time in our model is calculated based on the explicitly estimated queue length of each approach of the coordinated arterial road before the green light starts at each cycle. In addition, to achieve better coordination performance, not only the phases that provide right of way to coordinated directions but also the phases that provide the right of way to uncoordinated directions are considered during the optimization of phase sequences. The effectiveness and efficiency of the proposed method are validated via simulations on VISSIM. Experimental results illustrate that the proposed method can efficiently reduce the average delay, the average queue length, and the average number of stops per vehicle on arterial roads while not impacting the overall performance of the road network.

Siphons are a kind of special structural objects in a Petri net, and plays a key role in synthesizing a live Petri net controller for flexible manufacturing systems. In order to obtain a small size Petri net controller, this paper introduces the concept of a controllable siphon basis. It then proves that a live Petri net controller can be established by adding a control place and related arcs to each strict minimal siphon (SMS) in a controllable siphon basis. The initial markings of control places are determined by an integer linear program. The number of control places in the obtained controllers is the same as the number of SMSs in the controllable siphon basis, while the latter is no more than that of the activity places in a Petri net model. An algorithm for constructing a controllable siphon basis is proposed, and a new deadlock prevention policy based on it is established. A few examples are provided to demonstrate the proposed concepts and policy and used to compare them with the state-of-the-art methods.

Deadlock prevention is crucial to the modeling of flexible manufacturing systems. In the Petri net framework, deadlock prevention is often addressed by siphon-based control (SC) policies. Recent research results show that SC methods can avoid full siphon enumeration by using mixed integer programming (MIP) to greatly increase the computational efficiency so that it can be applied in large systems in computable time. Besides, maximally permissive control solutions can be obtained by means of iterative siphon control (ISC) approaches and MIP. Then the remaining problems are redundancy and MIP iterations. Redundant controllers make the closed-loop system more complicated and each MIP iteration increases the total computational time. This article proposes a revised ISC deadlock prevention policy which can achieve better results than the other reported methods in terms of redundancy and MIP iterations while maintaining the maximal permissiveness. Several benchmark examples are provided to illustrate the proposed approach and to be compared with the other reported methods.

This paper presents a novel method to design Petri net-based deadlock prevention controllers for flexible manufacturing systems. It starts from the computation of the complete deadlock markings by utilizing the conservativeness property of a Petri net model and the necessary and sufficient condition for deadlock. Then, it verifies a small state space including the dangerous and bad markings only by combining one-step look-forward through the original net and one-step look-backward via its reverse net. Subsequently, it defines the set of place invariants from the subset of marked operation places for the so called ¿elementary controlled bad markings¿. Finally, it synthesizes a deadlock prevention controller by a simplified invariant-based method. Its obtained deadlock-free controller allows more behavior of the closed-loop system than those obtained via a siphon-based control method. Its computational efficiency is higher than those based on a complete reachability graph-based control method.

The considered class of generalized mutual exclusion constraints (GMECs) on a controlled Petri net are such that the influence-uncontrollable subnets are forward-concurrent-free nets. Some structural properties of forward-concurrent-free nets are firstly proposed. Utilizing these properties, a method is then proposed to transform a given conjunction of GMECs into a conjunction of admissible GMECs. Furthermore, the necessary and sufficient condition of the existence of the permissive supervisor is obtained, and the optimal supervisor with a complexity of polynomial time is designed. The theoretic results are illustrated by an example that synthesizes a maximally permissive supervisor for a manufacturing system.

This paper addresses the state feedback control synthesis of discrete event systems on the forbidden state problem in which the forbidden states are finite, especially if they can not be expressed as linear inequality constraints using reported methods. The system is modelled by controlled Petri nets that ape bounded Petri nets or unbounded Petri nets with uncontrollable subnets satisfying the Reverse net Structurally Bounded Condition. Through the analysis of the reverse net, we obtain not only the weakly forbidden markings used to deal with uncontrollable transitions but also the maximally permissive state feedback control policy. Moreover, it is illustrated by an example in the reported literature that the method can be applied conveniently to a class of Petri nets whose uncontrollable subnets are Output Dominant Petri nets.

This paper addresses the enforcement of a class of linear inequality constraint defined on the marking of a Petri net (PN). The constraint may be regarded as the conjunction of 'less-than-or-equal-to' inequality and 'greater-than-or-equal-to' inequality. The extended Petri nets such as inhibitor arc PN and its complementary net, the so-called enabling arc PN, are exploited to design a PN supervisor such that the constraint is enforced in the controlled net. The supervisor is optimal in the sense that it allows the net evolves with least restriction while the given constraint is satisfied. An example is provided for illustration.

This paper addresses the supervisory control of the class of discrete event system (DES) modeled by a Petri net. The control specification described by a linear 'less-than-or-equal-to' inequality defined on the place marking of the net has been extensively studied in the literatures. However, in this paper we consider the control specification in the form of linear 'greater-than-or-equal-to' marking inequality. The supervisory control of the DES with 'greater-than-or-equal-to' constraint is implemented via an enabling arc, which is a recently proposed arc by Uzam (1998) and Uzam et al. (1999) and can be regarded as complementarity of inhibitor arc. An example illustrates the supervisory control method is presented in this paper.

A Petri net (PN) with weighted inhibitor arc is exploited to solve the forbidden state problem of discrete event systems (DES). The forbidden state problem considered is described as the linear inequality constraint of the place marking. We first review the relevant work on the control of DES using inhibitor arcs. Then, the design of the PN controller is introduced with two steps. The first step of the design is to track the state of the system. Then, the weighted inhibitor arcs are exploited to disable the relative transitions in the case that the firing of these transitions will violate the constraints. A simple example of a discrete manufacturing system in the reported literature is used to show the detailed procedure of the controller synthesis and the advantages of the presented method.

In the reported literatures on the control of discrete event systems (DES) modeled by Petri nets with the constraint of a logical intersection of some linear inequalities or just a single one, the constraint usually has to be transformed into the form of logical union when there are some uncontrollable transitions in the net. In this paper, we propose a hierarchical control method for DES with logical union of constraints based on colored Petri nets though the plant (uncontrolled DES) is modeled as a noncolored Petri net. The low-level gets the state information from the plant and sends it to the high-level. The high-level plays the role of controller and ensures that there is at least one of the constraints is satisfied at any time for any system state. In addition, it is proved that the proposed hierarchical control is maximal control.

The previous work of Boel et al. (1996) on the forbidden state problem for the class of discrete event systems modeled by controlled state graph is extended in this paper. The class of Petri nets we consider is a generalized net in which the weight of the arcs may exceed 1. Furthermore, the limit to the output places number of a transition in the net is removed and consequently the Petri net is capable of firing more than one process simultaneously. Based on the calculation of weakly forbidden conditions, the maximally permissive supervisor is obtained.