Multi-agent reinforcement learning (MARL) meth-ods face a curse of dimensionality in the policy and value function representations as the number of agents increases. The development of distributed or parallel training techniques is also hindered by the global coupling among the agent dynamics, requiring simultaneous state transitions. This paper introduces Distributed multi-Agent Reinforcement Learning with One-hop Neighbors (DARLIN). DARLIN is an off-policy actor-critic MARL method that breaks the curse of dimensionality and achieves distributed training by restricting the agent interactions to one-hop neighborhoods. Each agent optimizes its value and policy functions over a one-hop neighborhood, reducing the representation complexity, yet maintaining expressiveness by training with varying numbers and states of neighbors. This structure enables the key contribution of DARLIN: a distributed training procedure in which each compute node simulates the state transitions of only a small subset of the agents, greatly accelerating the training of large-scale MARL policies. Comparisons with state-of-the-art MARL methods show that DARLIN significantly reduces training time without sacrificing policy quality as the number of agents increases.

High-resolution optical cameras have always been important scientific payloads in Mars exploration missions, and the Mars topographic data produced by their detection data can provide support for scientific research on Mars geomorphology and geological structure evolution. As of December 2021, there are still relatively few high-resolution image data at the sub-meter level on the Martian surface, with about 2.6% global coverage and even more limited stereo coverage (just about 0.4%). At the same time, there are still some difficulties in data acquisition and terrain reconstruction processing methods for high-resolution Mars images that need to be solved. This paper described how we designed the in-orbit stereo imaging strategy based on the characteristics of the High Resolution Imaging Camera (HiRIC) of China’s first Mars exploration mission (Tianwen-1), studied the technical solutions for HiRIC stereo image photogrammetry processing, and produced a topographic dataset for the “Tianwen-1” landing area, including digital orthophoto map (DOM) with a GSD (ground sample distance) of 0.7, 3.5m, and digital elevation model (DEM) with a GSD of 3.5m. Precision analysis results show that this topographic data has good consistency in planar position and elevation compared with the existing Mars terrain data, and has advantages in spatial resolution and terrain detail expression, which will be widely used in the geological background study of the “Tianwen-1” landing area, as well as landing site positioning, Martian surface remote operation planning and other Mars scientific research and engineering tasks.

Federated learning (FL) is a promising machine learning paradigm because it allows distributed edge devices to collaboratively train a model without sharing their raw data. In practice, a major challenge to FL is that edge devices are heterogeneous, so slow devices may compromise the convergence of model training. To address such a challenge, several recent studies have suggested different solutions, in which a promising scheme is to utilize coded computing to facilitate the training of linear models. Nevertheless, the existing coded FL (CFL) scheme is limited by a fixed coding redundancy parameter, and a weight matrix used in the existing design may introduce unnecessary errors. In this paper, we tackle these issues and propose a novel framework, namely CFL-HC, to facilitate CFL in heterogeneous computing scenarios. In our framework, we consider a computing system consisting of a central server and multiple computing devices with original or coded datasets. Then we specify an expected number of input-output pairs that are used in one round. Within such a framework, we formulate an optimization problem to find the best deadline of each training round and the optimal size of the computing task allocated to each computing device. We then design a two-step optimization scheme to obtain the optimal solution. To evaluate the proposed framework, we develop a real CFL system using the message passing interface platform. Based on this system, we conduct numerical experiments, which demonstrate the advantages of the proposed framework, in terms of both accuracy and convergence speed.

Mobile ad hoc computing (MAHC), which allows mobile devices to directly share their computing resources, is a promising solution to address the growing demands for computing resources required by mobile devices. However, offloading a computation task from a mobile device to other mobile devices is a challenging task due to frequent topology changes and link failures because of node mobility, unstable and unknown communication environments, and the heterogeneous nature of these devices. To address these challenges, in this paper, we introduce a novel coded computation scheme based on multi-agent reinforcement learning (MARL), which has many promising features such as adaptability to network changes, high efficiency and robustness to uncertain system disturbances, consideration of node heterogeneity, and decentralized load allocation. Comprehensive simulation studies demonstrate that the proposed approach can outperform state-of-the-art distributed computing schemes.

This paper aims to mitigate straggler effects in synchronous distributed learning for multi-agent reinforcement learning (MARL) problems. Stragglers arise frequently in a distributed learning system, due to the existence of various system disturbances such as slow-downs or failures of compute nodes and communication bottlenecks. To resolve this issue, we propose a coded distributed learning framework, which speeds up the training of MARL algorithms in the presence of stragglers, while maintaining the same accuracy as the centralized approach. As an illustration, a coded distributed version of the multi-agent deep deterministic policy gradient (MADDPG) algorithm is developed and evaluated. Different coding schemes, including maximum distance separable (MDS) code, random sparse code, replication-based code, and regular low density parity check (LDPC) code are also investigated. Simulations in several multi-robot problems demonstrate the promising performance of the proposed framework.

In recent years, we have witnessed fast‐growing unmanned aerial systems (UAS) based applications. To better facilitate these applications, many efforts have been made to enhance the capability of UAS from various aspects, including communications, control and networking, and so on. Nevertheless, most of these studies neglect the computation aspect. Recently, the UAS‐enabled mobile edge computing (MEC) has attracted increasing research attention, which utilises UAS with onboard computing capability to provide on‐demand computing services for mobile users. However, existing research on UAS‐enabled MEC remains at the theory stage and how to design a UAS platform with advanced onboard computing capability has not been addressed. In this study, the authors aim to fill this research gap and design an open UAS‐based airborne computing platform with advanced onboard computing capability. This platform was designed from three aspects: hardware, software, and applications. In particular, feasible computing hardware to perform UAS onboard computing is first considered and a prototype is then designed. To enhance the flexibility and programmability of the platform, two key virtualisation techniques are then investigated. Finally, they test the performance of their prototype by executing real UAS onboard computing tasks, the results of which verify the feasibility and potentials of the proposed airborne computing platform.

Fitting primitives is of great importance for remote sensing applications, such as 3-D modeling and as-built surveys. This letter presents a method for fitting primitives that fuses the Bayesian sample consensus (BaySAC) algorithm with a statistical testing of candidate model parameters for unorganized 3-D point clouds. Instead of randomly choosing initial data sets, as in the random sample consensus (RANSAC), we implement a conditional sampling method, which is the BaySAC, to always select the minimum number of data required with the highest inlier probabilities. As the primitive parameters calculated by the different inlier sets should be convergent, this letter presents a statistical testing algorithm for the histogram of the candidate model parameter to compute the prior probability of each data point. Moreover, the probability update is implemented using the simplified Bayes formula. The proposed approach is tested with the data sets of planes, tori, and curved surfaces. The results show that the proposed optimized BaySAC can achieve high computational efficiency (five times higher than the efficiency of the RANSAC for fitting a subset of 12 500 points) and high fitting accuracy (on average, 20% higher than the accuracy of the RANSAC). Moreover, the strategy of prior probability determination is proven to be model-free and, thus, highly applicable.