The DETR series, known for its end-to-end object detection models, has gained significant attention for its performance. RT-DETR excels with higher accuracy and faster real-time inference. However, applying these models to medical imaging poses challenges, such as low-contrast and complex lesion structures, which can reduce effectiveness. When detecting tumors, models may overfit due to the distinct differences and variability between different cases, affecting generalization and accuracy. To address these challenges, we propose a multi-view parallel feature extraction module, specifically for tumor detection. This module includes adaptive preprocessing, joint axial and channel attention, multi-pooling angular attention to enhance relevant features and reduce redundancy. Additionally, axial dynamic deformable convolution is used to improve adaptability and robustness. The resulting PMA-DETR architecture achieves state-of-the-art tumor detection while maintaining real-time processing.

Global fixed-priority (G-FP) scheduling is a widely applied scheduling policy for real-time systems running on multiprocessor platforms. The state-of-the-art in priority assignment for G-FP follows one of two approaches. The first is to use a simple heuristic for priority assignment that works with any (thus the most accurate) schedulability analysis. The second is to leverage Audsley’s polynomial-time optimal priority assignment (OPA) algorithm, which can only accommodate a less accurate analysis that satisfies the compatibility conditions required by OPA. In this paper, we study this critical issue and present a novel algorithm. We first use the concept of response time estimation range to build a new priority assignment framework, which is optimal with a more accurate schedulability analysis than OPA since its compatibility conditions are much weaker than those of OPA. This new frontier on the second approach is then judiciously combined with the first approach to take advantage of both. We evaluate the effectiveness of the proposed algorithm with various task sets. Compared with existing approaches, our algorithm always achieves the highest acceptance ratio and can outperform them by 25% on average.

Emerging connected vehicle services powered by artificial intelligence and data analytic are gaining increasing interest and attention with the advancement of cloud/edge computing technologies. Given the highly data- and computation-intensive characteristics of these applications, it is important that requests for these services be carefully allocated in cloud/edge computing systems to optimize performance, resource utilization and cost. Challenges arises when mobility of vehicles is taken into consideration. Specifically, as services become increasingly sophisticated and computation intensive, a vehicle may travel non-trivial amount of distance during queuing and processing of a request, which affect transmission of result data upon service fulfillment. In these cases, it is important that allocation of request be aware of the expected position of the vehicle at the time of request completion as oppose to submission. In the general scenario where there exists multiple cloud/edge devices and resource contention, it is important to simultaneously consider vehicle trajectories, workload and scheduling of requests to jointly optimize allocation. In this paper, we study the problem of cost minimization in allocation and scheduling of connected vehicle service requests on heterogeneous cloud/edge services. We consider the scenario where vehicles have non-trivial mobility during service delay and model its impact on data transmission. We introduce an optimal ILP formulation as well as an efficient and close to optimal heuristic algorithm for solving the optimization problem. Experiment result shows that the proposed technique is capable of achieving 10% to 30% of improvement comparing with straightforward approaches.

The design of modern real-time systems not only needs to guarantee their timing correctness, but also involves other critical metrics such as control quality and energy consumption. As real-time systems become increasingly complex, there is an urgent need for efficient optimization techniques that can handle large-scale systems. However, the complexity of schedulability analysis often makes it difficult to be directly incorporated in standard optimization frameworks, and inefficient to be checked against a large number of candidate solutions. In this paper, we propose a novel optimization framework for the design of real-time systems. It leverages the sustainability of schedulability analysis that is applicable for a large class of real-time systems. It builds a counterexample-guided iterative procedure to efficiently learn from an unschedulable solution and rule out many similar ones. Compared to the state-of-the-art, the proposed framework may be ten times faster while providing solutions with the same quality.

In cyber-physical systems, certain tasks are activated according to a rotation source. For example, angular tasks in engine control units are triggered whenever the engine crankshaft reaches a specific angular position. To reduce the workload at high speeds, these tasks also adopt different implementations within different rotation speed intervals. However, current studies are limited to the case that the switching speeds at which task implementations should change are configured at design time. In this article, we propose to dynamically adjust the switching speeds at runtime. We develop schedulability analysis techniques for such systems, including a new digraph-based task model to safely approximate the workload from software tasks triggered at predefined rotation angles. We prove that such task transformation has bounded pessimism. We present exact algorithms to find a finite number of representatives to avoid enumerating (an infinite number of) all job sequences. Experiments on synthetic task systems demonstrate that the proposed approach provides substantial benefits on system schedulability.

Today's automotive engine control systems adopt several control strategies that come with tradeoffs between computational load and performance. The current practice is that the switching speeds at which the engine control system changes control strategy is fixed offline, typically based on the average driving need in a standard driving cycle (i.e., vehicle speed profile over time). This is clearly suboptimal since it fails to capture the variation in the driving cycle, and the actual driving cycle may be considerably different from the standard one. In this paper, we propose to dynamically adjust switching speeds based on the predicted driving cycle. We develop a hybrid set of schedulability analysis techniques to tame the complexity of ensuring the real-time schedulability of engine control tasks. We design an effective and efficient optimization algorithm that provides close-to-optimal solutions. Experimental results demonstrate that our approach efficiently finds dynamic switching speeds that significantly improve engine performance over static ones.

In the design optimization of real-time systems scheduled with fixed priority, schedulability analysis is used to define the feasibility region within which tasks meet their deadlines, so that optimization algorithms can find the best solution within the region. However, the complexity of schedulability analysis techniques often makes it difficult to leverage existing optimization frameworks and scale to large designs. In this paper, we propose the concept of unschedulability core, a compact representation of the schedulability conditions, and develop efficient algorithms for its calculation. We present a new optimization framework that leverages such a concept. We show that this concept is applicable to a range of optimization problems, for example, when the decision variables include the task priority assignment and the selection of mechanisms protecting shared buffers. Experimental results on two case studies demonstrate that the new optimization procedure maintains the optimality of the solutions, but is a few orders of magnitude faster than other exact algorithms (branch-and-bound, integer linear programming).

Priority assignment is a critical issue in the design of real-time systems scheduled with fixed priority. In this paper, we consider the design optimization of a class of such real-time systems, where the schedulability of a task not only depends on the set of higher/lower priority tasks, but also on the response times of other tasks. This makes Audsley's algorithm inapplicable for finding a schedulable priority assignment, not to mention that the design optimization may also involve other metrics such as end-to-end latency in the constraints or objective. The current approaches are to either develop heuristics or use standard exhaustive search algorithms such as Branch-and-Bound (BnB). Instead, we propose a new approach that breaks through the mindset of the current approaches. Our main idea is to introduce the concept of response time estimation range, which is used in place of the actual response time of each task for evaluating the system schedulability. This allows to leverage Audsley's algorithm to generalize from an unschedulable solution to many similar ones. We develop an optimization framework that builds upon such a concept. We then apply the proposed approach to industrial designs of two use cases. One is the real-time wormhole communication in a Network-on-Chip (NoC), where a traffic flow suffers indirect interferences that depend on the response times of higher priority flows. The other is distributed systems with data-driven activation, where a task is triggered by the availability of data, hence by the completion of its immediate predecessor. Experimental results show the proposed technique typically runs several times faster than exhaustive search algorithms based on BnB or Mixed Integer Linear Programming (MILP).

Model-based design using the Simulink modeling formalism and associated toolchain has gained popularity in the development of real-time embedded systems. However, the current research on software synthesis for Simulink models has a critical gap for providing a deterministic, semantics-preserving implementation on multicore architectures with partitioned fixed-priority scheduling. In this paper, we consider a semantics-preservation mechanism that combines (1) the RT blocks from Simulink, and (2) task offset assignment to separate the time windows to access shared buffers by communicating tasks. We study the software synthesis problem that optimizes control performance by judiciously assigning task offsets, task priorities, and task communication mechanisms. We develop a problem-specific exact algorithm that uses an abstraction layer to hide the complexity of timing analysis. Experimental results show that it may run a few orders of magnitude faster than a direct formulation in integer linear programming.

In real-time embedded systems certain tasks are activated according to a rotation source, such as angular tasks in engine control unit triggered whenever the engine crankshaft reaches a specific angular position. To reduce the workload at high speeds, these tasks also adopt different implementations at different rotation speed intervals. However, the current studies limit to the case that the switching speeds at which task implementations should change are configured at design time. In this paper, we propose to study the task model where switching speeds are dynamically adjusted. We develop schedulability analysis techniques for such systems, including a new digraph-based task model to safely approximate the workload from software tasks triggered at predefined rotation angles. Experiments on synthetic task systems demonstrate that the proposed approach provides substantial benefits on system schedulability.

Modern embedded systems, such as automotive, are physically distributed with an increasing number of microcontrollers and buses. They support complex functions such as active safety and autonomous driving features with a high degree of data dependencies. The most common configuration uses periodic activation of tasks and messages coupled with priority-based scheduling. Selecting task and message parameters so that end-to-end deadlines are met can be very challenging, since such deadlines are enforced across a set of microcontrollers and buses. In this paper, we address the problem of optimal selection of task and message activation periods and priorities. Existing approaches cannot scale to large designs and have to settle to optimize period or priority separately, largely due to the complexity of response time analysis techniques. Instead, we present a new, unified framework that simultaneously optimizes period and priority assignment. It avoids the pitfalls of existing approaches by abstracting the response time calculation with the new concept of maximal unschedulable period and deadline assignment. We demonstrate with two industrial case studies that our approach runs magnitudes faster than existing approach on period optimization, while providing substantially better solutions.

In the design optimization of real-time systems, the schedulability analysis is used to define the feasibility region within which tasks meet their deadlines, so that optimization algorithms can find the best solution within the region. However, the complexity of current schedulability analysis techniques often makes it difficult to leverage existing optimization frameworks and scale to large designs. In this paper, we consider the design optimization problems for real-time systems scheduled with fixed priority, where task priority assignment is part of the decision variables. We propose the concept of unschedulability core, a compact representation of the schedulability conditions, and develop efficient algorithms for its calculation. We present a new optimization procedure based on lazy constraint paradigm that leverages such a concept. Experimental results on two case studies show that the new optimization procedure provides optimal solutions, but is a few magnitudes faster than other exact algorithms (Branch-and-Bound, Integer Linear Programming).

This paper considers the problem of design optimization for real-time systems scheduled with fixed priority, where task priority assignment is part of the decision variables, and the timing constraints and/or objective function linearly depend on the exact value of task response times (such as end-to-end deadline constraints). The complexity of response time analysis techniques makes it difficult to leverage existing optimization frameworks and scale to large designs. Instead, we propose an efficient optimization framework that is three magnitudes (1,000×) faster than Integer Linear Programming (ILP) while providing solutions with the same quality. The framework centers around three novel ideas: (1) An efficient algorithm that finds a schedulable task priority assignment for minimizing the average worst-case response time; (2) The concept of Maximal Unschedulable Deadline Assignment (MUDA) that abstracts the schedulability conditions, i.e., a set of maximal virtual deadline assignments such that the system is unschedulable; and (3) A new optimization procedure that leverages the concept of MUDA and the efficient algorithm to compute it.

Model-based design using Synchronous Reactive (SR) models is becoming widespread for control software development in industry. However, software synthesis is challenging for multi-rate SR models consisting of blocks modeled with finite state machines, due to the complexity of validating the system’s real-time schedulability. The existing approach uses the simplified periodic task model to allow an efficient schedulability analysis, which leads to pessimistic and suboptimal solutions. Instead, in this paper, we adopt a more accurate but more complex schedulability analysis. We develop several optimization techniques to improve the algorithm’s efficiency. Experimental results on synthetic systems and an industrial case study show that the proposed optimization framework preserves the solution optimality but is much faster (e.g., 1000× for systems with 15 blocks) than the branch-and-bound algorithm, and it generates better control software than the existing approach.