The deployment of mobile robotic systems to carry out in-house transportation tasks is a key element to improve efficiency in the logistics of Industry 5.0-inspired smart factories. However, the standard requirements of logistics schemes such as flexibility, reconfigurability, reusability, scalability or energy-efficiency, pose a number of challenging open –from the optimality side– control problems to be addressed. Essentially, the main issues to by faced are related to scheduling or dispatching, which encompasses task assignment and empty vehicle balancing, and routing. This is to be done under well-defined optimization criteria and taking into account the specific features of the Automated Guided Vehicle (AGV) fleet, both static (size, load), kinematic (velocity, acceleration), and technical (holonomicity, autonomy, sensing and interconnection capabilities). Additional constraints have to do with the degree of heterogeneity of the fleet, and the level of interaction with human operators. Most of the solutions reported so far consider centralized control strategies, where a central unit makes decisions relying on global information of the whole system. However, depending on the number of units the communication overhead might have a negative impact on performance, while the control complexity often results in NP-hard computational problems. A promising alternative that is gaining an increasing interest in the specialized literature comes from distributed control systems. Decision-making algorithms based on local information require less individual computational power and are more robust to disturbances, while its intrinsic modular structure offers excellent flexibility and scalability properties. The thematic area of this thesis proposal falls within the traffic management of AGV fleets in in-house facilities using distributed control approaches.
Multi-agent systems are at the core of the decentralized approaches to tackle the task assignment problem in conventional environments. The cooperation capabilities via intercommunication allow them to pursue for collective targets that encompass both individual goals and global behaviors. These features are exploited by the so-called market-based algorithms, in which assignments come through an auction based on certain criteria. Such auctions may occur either for neighboring units of the loading station or within all the fleet, and AGVs may bid to different station offers.
Empty vehicle balancing deals with the managing of idle vehicles, including the assignment of AGVs to battery charging stations. Intuitively, parking/charging stations and vehicles should be appropriately balanced between different zones so as to minimize response times. Circulatory loop positioning or even random motion through the warehouse of idle AGVs are also alternatives for short response times, but it also shortens individual intercharge times. Some works also include this operation within the task assignment routines.
Routing or path planning is an essential computation to be taken into account at the task assignment stage. Modern route planning strategies work on a dynamic basis, i.e. taking into account the current state of the network to avoid congestions, and assign routes using consensus-based approaches between AGV neighbors that minimize the global execution time of a set of tasks. Once the route is selected, route execution algorithms are in charge of operationally driving the AGV to its destination avoiding collisions and deadlock situations. A key difference is also on the existence of specific moving lanes; when not, the AGVs are classified as free ranging vehicles, which offer better space utilization and flexibility at the cost of algorithm complexity.
Despite the current advances in the area, further research is needed on the integration of scheduling and route planning processes, which are often analyzed separately, as it is a key element in distributed strategies. An appropriate balancing of the level of decentralization in distributed decision-making processes is also desirable, as a reduction of the amount of non-local information managed entails less computational and communication power effort and, at the same time, improves robustness, flexibility, and scalability. Moreover, market-based strategies in task assignment could be ameliorated with local optimization algorithms and take into account side constrains including space, time, capacity, battery level… Another important point is the development of strategies dealing with the simultaneous scheduling of multiple types of materials and heterogeneous AGV fleets, i.e. where different AGV models coexist. Route execution processes would benefit of allowing AGVs to eventually leave moving lanes in order to better sort out collision-avoidance and deadlock situations. As for deadlock prevention, low rule-based strategies require less computational effort. Finally, the ability of algorithms to work in mixed scenarios, with different control paradigms, is also a challenging task.
The thesis will be focused on the development of distributed control strategies for the traffic management of AGV-based in-house transportation systems encompassing:
• the task scheduling and route planning of the fleet in an integrated fashion and with a high level of decentralization,
• the route execution of individual AGVs during operation in potentially mixed scenarios with improved collision avoidance and deadlock properties.