Conference Agenda

Modern robotics applications increasingly demand heterogeneous multi-robot fleets to simultaneously handle diverse, specialized tasks such as collaborative material transport, manipulation, and sorting that single-purpose automation systems cannot manage alone. However, executing these complex logistical processes and scientific experiments requires advanced process control systems capable of managing highly dynamic and intricate workflows. While existing fleet management solutions often lack the flexibility to handle diverse robotic capabilities, this paper introduces a comprehensive process control concept specifically designed for the orchestration and monitoring of diverse robotic agents. This concept directly addresses the challenge of dynamically coordinating interdependent, multi-robot tasks without relying on hardcoded workflows. Rather than focusing on the underlying software architecture, the primary emphasis is placed on the core order-modeling and control mechanisms that enable the seamless definition and coordination of complex, multi-robot workflows.

At the heart of the proposed concept is a highly flexible order definition framework. The foundational order structure is broadly oriented toward the VDA 5050 standard, adopting core concepts like nodes, edges, and actions for order creation, transmission, and monitoring. To meet the demands of advanced robotic operations, such as tightly coupled multi-robot collaboration, real-time reactive behaviors, and environment-dependent task execution, this structure is significantly extended by a suite of powerful control mechanisms. These include flexible start and end conditions, spatial zones for coordination, and the use of runtime placeholders. These placeholders allow action parameters to be dynamically assigned during execution using real-time data or results from other system components.

To manage the temporal and logical dependencies between multiple robots, the concept incorporates a dedicated cross-order coordination backbone. This framework supports synchronization actions, such as sync lists and specific triggers, as well as the capacity for repeating orders that loop dynamically based on defined events, robot states, or manual user confirmations via a graphical interface. The capabilities of this process control framework are demonstrated through complex, multi-robot use cases, such as the collaborative transport of large loads and the boarding process of a delivery robot into an autonomous shuttle.

Finally, the paper addresses current limitations regarding the ongoing expansion efforts of existing standards and outlines future conceptual pathways to scale the process control system for next-generation robotic research environments.

Urban railways are critical logistics systems that move large passenger flows through time-sensitive and capacity-constrained networks. When disruptions occur, their effects can propagate rapidly across stations, trains, operators, and passenger journeys, requiring coordinated decisions among operations, maintenance, and customer-service teams. For advanced logistics management, disruption records should therefore support not only retrospective reporting but also evidence-based severity assessment, operational learning, and future integration with simulation or digital-twin environments. However, in Thailand, rail disruption narratives are often recorded as mixed Thai–English free text with inconsistent spacing, abbreviations, and operator-specific expressions, limiting the usefulness of structured incident codes alone for large-scale analytics.

This study proposes a machine-learning-based natural language processing workflow for classifying disruption severity using daily records from the Department of Rail Transport dataset during 2021–2025 (N = 770). Severity labels were deterministically derived from delay minutes and grouped into three operational classes: A for delays below 5 minutes, B for delays from 5 to below 15 minutes, and S for delays of 15 minutes or more. These labels are interpreted as a proxy for passenger and operational impact. The DETAIL field was used as the main narrative input. Text preprocessing included removal of repeated spaces and Thai word segmentation using PyThaiNLP with a custom dictionary developed from station names, station codes, and disruption-cause terms to improve recognition of railway-specific expressions in mixed Thai–English text.

The experiment compared four text feature representations: Bag-of-Words, TF-IDF, Trigram features, and WangchanBERTa sentence embeddings. Each representation was evaluated with Logistic Regression and XGBoost. Two validation settings were used. A stratified split divided the data into 70% training, 20% validation, and 10% testing while preserving severity-class proportions. A chronological split used the same proportion but sorted records by date to evaluate temporal generalization for future deployment.

The dataset was imbalanced, with Class S accounting for 429 records (55.71%), Class B for 268 records (34.81%), and Class A for 73 records (9.48%). The average delay was 30.02 minutes, with a median of 15 minutes and a maximum of 4,320 minutes. Under the stratified split, BoW + XGBoost achieved the best balanced performance, with accuracy = 0.870 and F1-score = 0.859. BoW + Logistic Regression also performed strongly, with accuracy = 0.844 and F1-score = 0.838. Under the chronological split, TF-IDF + XGBoost achieved the highest accuracy of 0.857, while WangchanBERTa embeddings + XGBoost achieved the highest F1-score of 0.607. However, the chronological test set did not contain Class A examples; therefore, chronological results must be interpreted carefully and should be considered together with the stratified-split results.

These findings show that mixed Thai–English disruption narratives contain meaningful severity signals and that lightweight, interpretable NLP models can provide a practical baseline for disruption monitoring, severity validation, escalation support, and future rail logistics decision-support to an operator or digital-twin applications.

Ports are central nodes in multimodal transport networks, linking maritime or inland waterway transport with rail and road-based hinterland connections. Within these nodes, rail shunting operations are operationally critical but difficult to plan digitally because transport orders, infrastructure states, locomotive availability, wagon movements, and timing information are often distributed across heterogeneous systems. Existing digital platforms in port logistics primarily support information exchange and process visibility, while less attention is given to the transformation of local operational states into executable shunting decisions under time, resource, and infrastructure constraints.

This paper presents a reference architecture developed within the KIRBI project, which investigates AI-supported shunting and dispatching processes in port rail operations. The architecture is currently being developed and prototyped in two German port railway contexts, Dortmunder Hafen and Hafen Hamm, and targets the transition from conceptual design toward prototypical implementation. The contribution focuses on the system-level integration required to couple operational data, digital infrastructure representation, analytical services, heuristic optimization, and dispatcher-oriented decision support.

The architecture follows a modular, event-driven design. Heterogeneous inputs such as transport orders, telematics or GPS data, train arrival notifications, master data on locomotives and wagons, infrastructure information, and camera-based observations are harmonized into a common operational state model. This state model represents the current planning situation, including available resources, pending tasks, relevant infrastructure elements, operational dependencies, and timing constraints. Analytical services then provide decision-relevant parameters for dispatching. In this context, AI-based components are used as supporting services, for example for estimating travel or availability times and for camera-based validation of wagon sequences. By replacing static planning assumptions with situation-dependent estimates, these services allow the heuristic optimizer to react to changing operational conditions such as delayed arrivals, varying resource availability, or deviations in wagon sequences.

The dispatching decision itself is generated by a heuristic optimization component. The current prototype uses a constructive planning logic with iterative improvement: initial locomotive-task sequences are derived from operational priorities and feasibility constraints and are subsequently refined with neighborhood-based adjustments. The optimizer considers task readiness, due times, shift structures, setup times, travel times, resource availability, operational dependencies, and return-to-base requirements. Generated proposals are presented through a decision-support interface where dispatchers can review, adjust, or reject suggested plans. Changed operational states, dispatcher feedback, and execution events can trigger repeated replanning and create a basis for future model calibration and empirical evaluation.

The architecture differs from generic port information systems by focusing on local, time-critical rail shunting decisions rather than inter-organizational information exchange alone. Its transferability is addressed through a separation between site-specific elements, such as infrastructure topology and local process rules, and reusable architectural components, such as data adapters, event-based communication, operational state modeling, analytical services, and heuristic dispatching logic. As the current development stage lies between concept and prototype, the paper does not claim quantitative performance improvements. Instead, it defines the architectural design rationale and an evaluation framework for subsequent empirical assessment, including indicators such as delay, empty movements, resource utilization, plan stability, dispatcher workload, and replanning responsiveness.