Managing actions with uncertain resource costs is a complex challenge, particularly in autonomous robot mission planning. Robots are often assigned multiple objectives with varying criticality levels, ranging from catastrophic to minor impacts, where failures can significantly affect system safety. Uncertainties in worst-case costs of resources, such as energy and operating time - the time it takes to carry out an action - further complicate mission planning and execution. Monte Carlo Tree Search (MCTS) is a powerful tool for online planning, yet it struggles to account for uncertainty in worst-case cost estimations. Optimistic estimates risk resource shortages, while pessimistic ones lead to inefficient allocation. The Mixed-Criticality (MC) approach, originally developed for real-time systems to schedule critical tasks by allocating processing resources under Worst-Case Execution Time (WCET) uncertainty, provides a framework of rules, models and design principles. We claim this framework can be adapted to autonomous robot mission planning, where critical objectives are met through analogous allocation of different kinds of resources such as energy and operating time despite uncertainties. We propose enhancing MCTS with MC principles to handle uncertainty in worst-case costs across multiple resources and criticality of objectives. High-critical objectives must always be completed, regardless of resource constraints, while low-critical objectives operate flexibly, consuming resources within optimistic estimates when possible or being discarded when resources become scarce. This ensures efficient resource reallocation and prioritization of high-critical objectives. To implement this, we present (MC)²TS, a novel variant of MCTS that integrates MC principles for dynamic resource management. It supports more than two criticality levels to ensure that the most critical components meet the most stringent safety and reliability requirements, while also enabling robust resource management. By enabling replanning and mode changes, (MC)²TS improves MCTS’s efficiency and enhances MC systems’ adaptability to both degrading and improving resource conditions. We evaluate (MC)²TS in an active perception scenario, where a drone retrieves data from distributed sensors under unpredictable environmental conditions. (MC)²TS outperforms MCTS by achieving more objectives, adapting plans when costs drop. It explores more objective sequences, minimizes oversizing, and enhances efficiency. Balancing safety and performance, it monitors robot battery, mission and objective resource constraints such as deadlines. Its robustness ensures low-critical objectives do not compromise high-critical objectives, making it a reliable solution for complex systems characterized by uncertain resource costs and critical objectives.
The Slot-Shifting algorithm presents a solution to combine the benefits of offline and online scheduling in time-triggered systems. It dynamically adjusts the allocation of time slots to tasks in the scheduling tables to accommodate aperiodic tasks at runtime. In this note, we revisit an extension to Slot-Shifting that enables it to handle sporadic task sets. In particular, we clarify the assumptions required for the correct application of its offline acceptance test, identify sources of pessimism within it, and address its schedulability analysis interval.
Elastic scheduling provides a framework under which the utilizations of recurrent tasks are reduced by increasing their periods in response to system overload. The original elastic scheduling model was proposed by Buttazzo et al. in 1998 for implicit-deadline tasks on a uniprocessor and decreases task utilizations to satisfy a schedulable utilization bound. In 2019, Orr and Baruah extended the framework to multiprocessor scheduling of implicit-deadline tasks. In this paper, we propose, analyze, and evaluate new elastic scheduling algorithms for several of the scheduling policies considered in these prior works. In particular, (i) we evaluate an algorithm that we proposed as a short note in the Real-Time Systems journal and demonstrate that it allows for faster admission control than the algorithm of Buttazzo et al. when applied to uniprocessor and fluid scheduling. (ii) We also present faster elastic scheduling algorithms for partitioned EDF scheduling. Finally, (iii) we provide polynomial-time exact elastic scheduling algorithms for global EDF and global RM.
The Asynchronous Traffic Shaping (ATS) has been designed by the Time Sensitive Networking (TSN) group as a reshaping mechanism for real-time data flows, based on the initial proposition of the Urgency Based Shaper (UBS). Several studies have exhibited properties and limitations of this solution, but most of them are based on the model presented in the UBS definition [Specht and Samii, 2016], whereas the implementation described in the standard uses a different architecture and algorithm. This paper presents an equivalence proof between the model and the standard specification.
Face recognition approaches often rely on equal image resolution for verifying faces on two images. However, in practical applications, those image resolutions are usually not in the same range due to different image capture mechanisms or sources. In this work, we first analyze the impact of image resolutions on face verification performance with a state-of-the-art face recognition model. For images synthetically reduced to 5 × 5 px resolution, the verification performance drops from 99.23% increasingly down to almost 55%. Especially for cross-resolution image pairs (one high- and one low-resolution image), the verification accuracy decreases even further. We investigate this behavior more in-depth by looking at the feature distances for every 2-image test pair. To tackle this problem, we propose the following two methods: 1) Train a state-of-the-art face-recognition model straightforwardly with 50% low-resolution images directly within each batch. 2) Train a siamese-network structure and add a cosine distance feature loss between high- and low-resolution features. Both methods show an improvement for cross-resolution scenarios and can increase the accuracy at very low resolution to approximately 70%. However, a disadvantage is that a specific model needs to be trained for every resolution pair. Thus, we extend the aforementioned methods by training them with multiple image resolutions at once. The performances for particular testing image resolutions are slightly worse, but the advantage is that this model can be applied to arbitrary resolution images and achieves an overall better performance (97.72% compared to 96.86%). Due to the lack of a benchmark for arbitrary resolution images for the cross-resolution and equal-resolution task, we propose an evaluation protocol for five well-known datasets, focusing on high, mid, and low-resolution images.
