Wireless power transfer enhanced mobile edge computing (WPT-MEC) has emerged as a key technology to support low-latency and energy-efficient computation in wireless networks. With increasing network density, multi-base-station architectures emerge where wireless devices (WDs) offload tasks to distributed base stations (BSs), creating challenges in maintaining quality-of-service fairness during complex resource coordination in multi-BS WPT-MEC networks. To address these challenges, we investigate a non-orthogonal multiple access (NOMA)-enhanced WPT-MEC network comprising multiple WDs and BSs with finite computational capacities. For ensuring fairness, we formulate a max-min problem to maximize the minimum task computation amount by jointly optimizing offloading decisions, NOMA decoding orders, offloading powers and time resource allocation, which results in a challenging mixed integer, sequence and nonlinear programming (MISNLP). To tackle this problem, we propose a two-stage distributed multi-agent algorithm. In the first stage, each BS agent generates offloading preferences based on partial observations, guiding WDs' offloading decisions. In the second stage, given these offloading decisions, we develop an efficient convex-based algorithm to solve the per-BS resource allocation subproblem, jointly optimizing NOMA decoding order, offloading powers and time resource allocation. For effective training, we leverage off-policy training and the centralized training with decentralized execution (CTDE) paradigm with two key innovations: (1) a convex-based critic that evaluates the joint action without bias, and (2) a counterfactual baseline that isolates individual agent credit assignment. The proposed C3MA algorithm achieves six times faster convergence and at least 20% performance improvement when serving more than 20 WDs, compared with existing multi-agent schemes, while maintaining a near-optimal Jain's fairness index of 0.97. Moreover, it sustains an ultra-low execution delay below 5 milliseconds even with 40 WDs, confirming its efficiency and scalability.

Mobile-edge computing (MEC) and wireless power transfer (WPT) are pivotal for enhancing computational power and battery life in 5G/6G networks. However, their performance declines in remote or disaster-stricken areas due to the lack of access points and energy sources. This paper proposes a wirelessly powered unmanned aerial vehicle enabled MEC (UAV-MEC) system to address this issue, focusing on nodes with ignorable computing capabilities and randomly arriving, size-varying tasks. We aim to maximize the long-term average computation rate under constraints such as UAV coverage, time resources, energy, and task causality, formulating a non-convex problem with dynamic states and complex actions. To solve this problem, we introduce an exploration-enhanced deep reinforcement learning (EDRL) algorithm with a bi-layered structure: the main problem determines the UAV’s flying actions, while the sub-problem allocates time resources given these actions. EDRL employs a deep neural network to analyze real-time UAV positions and task demands, determining optimal flight paths. Upon path determination, an efficient algorithm utilizing bisection and golden section search methods allocates WPT and computational offloading durations. Simulations reveal that EDRL achieves an execution latency of just 11.5 ms in thirty-node networks, outperforming baseline DRL algorithms and predetermined trajectory schemes by 20% and 25% in long-term computation rates, respectively. These results highlight EDRL’s effectiveness and low computational complexity, making it a robust solution for challenging environments.

Unmanned Aerial Vehicle (UAV)-enabled wirelessly powered Mobile Edge Computing (MEC) is emerging as a powerful technology for boosting computational capability and energy supplementation in Internet of Things (IoT). This work addresses the long-term task completion ratio maximization problem in UAV-enabled wirelessly powered MEC systems. Besides the large number of optimization parameters, the environment can only be partially observed as the UAVs cannot cover the whole network area. Then, it is very challenging to obtain good solutions due to the lack of global information. We introduce a novel distributed Multi-Agent Deep Reinforcement Learning (MADRL) framework for optimizing UAVs’ actions and resource allocation, considering the constraints of tasks that vary in size, arrival times, and required computation completion time. To decouple the complicated parameters, we divide the problem into two manageable subproblems—UAVs’ action decision and resource allocation under a given UAV’s action. We employ a distributed Deep Reinforcement Learning (DRL) scheme for the former subproblem to cope with the partially observable nature. By revealing some important properties of the later subproblem, we design an efficient two-stage optimal algorithm to minimize the total consumed energy of nodes while maximizing the task-completing number. Extensive simulations validate the effectiveness of the proposed framework, achieving over a 50% improvement in task completion ratio compared to baseline schemes in some scenarios.

This paper presents a method for identifying mechanical parameters of robots or objects, such as their mass and friction coefficients. Key features are the use of off-the-shelf physics engines and the adaptation of a Bayesian optimization technique towards minimizing the number of real-world experiments needed for model-based reinforcement learning. The proposed framework reproduces in a physics engine experiments performed on a real robot and optimizes the model's mechanical parameters so as to match real-world trajectories. The optimized model is then used for learning a policy in simulation, before real-world deployment. It is well understood, however, that it is hard to exactly reproduce real trajectories in simulation. Moreover, a near-optimal policy can be frequently found with an imperfect model. Therefore, this work proposes a strategy for identifying a model that is just good enough to approximate the value of a locally optimal policy with a certain confidence, instead of wasting effort on identifying the most accurate model. Evaluations, performed both in simulation and on a real robotic manipulation task, indicate that the proposed strategy results in an overall time-efficient, integrated model identification and learning solution, which significantly improves the data-efficiency of existing policy search algorithms.

This paper aims to identify in a practical manner unknown physical parameters, such as mechanical models of actuated robot links, which are critical in dynamical robotic tasks. Key features include the use of an off-the-shelf physics engine and the Bayesian optimization framework. The task being considered is locomotion with a high-dimensional, compliant Tensegrity robot. A key insight, in this case, is the need to project the space of models into an appropriate lower dimensional space for time efficiency. Comparisons with alternatives indicate that the proposed method can identify the parameters more accurately within the given time budget, which also results in more precise locomotion control.

Precise facial feature points extraction is essential to the high-level face recognition and expression analysis. This paper presents a method which uses Active Appearance Models with Inverse Compositional Lucas-Kanade Image Alignment algorithm for facial feature points extraction. Experimental results show that the presented method is capable of extraction of precise facial feature points from face. The feature points extraction is robust against the expression changes and scale variation.