In this paper, we propose a novel Deep Reinforcement Learning (DRL) algorithm which can navigate non-holonomic robots with continuous control in an unknown dynamic environment moving obstacles. We call the approach MK-A3C (Memory and Knowledge-based Asynchronous Advantage Actor-Critic) for short. As its first component, builds GRU-based memory neural network to enhance robot's capability temporal reasoning. Robots without it tend suffer from lack of rationality face incomplete noisy...

Reinforcement learning (RL)-based traffic signal control has been proven to have great potential in alleviating congestion. The state definition, which is a key element RL-based control, plays vital role. However, the data used for definition literature are either coarse or difficult measure directly using prevailing detection systems control. This paper proposes deep reinforcement learning-based method uses high-resolution event-based data, aiming achieve cost-effective and efficient...

Since an individual approach can hardly navigate robots through complex environments, we present a novel two-level hierarchical framework called JPS-IA3C (Jump Point Search improved Asynchronous Advantage Actor-Critic) in this paper for robot navigation dynamic environments continuous controlling signals. Its global planner JPS+ (P) is variant of JPS Search), which efficiently computes abstract path neighboring jump points. These nodes, are seen as subgoals, completely rid Deep Reinforcement...

In this paper, we present a hierarchical path planning framework called SG-RL (subgoal graphs-reinforcement learning), to plan rational paths for agents maneuvering in continuous and uncertain environments. By "rational", mean (1) efficient eliminate first-move lags; (2) collision-free smooth with kinematic constraints satisfied. works two-level manner. At the first level, uses geometric path-planning method, i.e., Simple Subgoal Graphs (SSG), efficiently find optimal abstract paths, also...

In the context of robotics and game AI, grid-based Distance Maps (DMs) are often used to fulfill collision checks by providing each traversable cell maximal clearance its closest obstacle. A key challenge for DMs’ application is how improve efficiency updating distance values when states changed (i.e., changes caused newly inserted or removed obstacles). To this end, paper presents a novel algorithm speed up construction DMs on planar, eight-connected grids. The novelty our algorithm,...

Exploration in unknown environments using deep reinforcement learning (DRL) often suffers from sampling inefficiency due to notoriously sparse extrinsic rewards and complex spatial structures. To this end, we present a hierarchical modular exploration model that integrates the recently popular concept of intrinsic motivation (IM). The approach addresses problem two levels. On higher level, DRL based global module learns determine distant but easily reachable target maximizes current...

When agents need to collaborate without previous coordination, the multi-agent cooperation problem transforms into an ad hoc teamwork (AHT) problem. Mainstream research on AHT is divided type-based and type-free methods. The former depends known teammate types infer current type, while latter does not require them at all. However, in many real-world applications, complete absence sufficient knowledge of are both impractical. Thus, this focuses challenge with limited types. To end, paper...

Motivation: The synthetic training data used for mapping reconstruction of deep learning can simulate the required features, but it is difficult to fully unnecessary characteristics existing in real-world data. Goal(s): This work aims enable model extract essential relationships and eliminate interference non-ideal factors real Approach: We propose a mask pre-training method called Masked U-net that allows learn appropriate inductive biases on quantitative images. Results: proposed better...

In this paper, a joint source and relay optimization problem is studied for multiuser multiple-input multiple-output (MIMO) system. Assuming that the channel state information (CSI) at available, two amplify forward (AF) relaying schemes are proposed under criterion of maximizing sum-rate. First, scheme which iteratively searches optimal matrices by deriving partial derivatives sum-rate applying gradient search algorithm proposed. Next, in order to reduce computational complexity, an...

Due to the mechanical failure data is cumulatively acquired and has uncertain features, memory model for fault diagnosis required adapt with information updating. In this paper, a dynamic using one-class support vector (OCSVM) proposed extract keep diagnostic information. The feature of each type respectively processed by incremental learning algorithm OCSVM construct optimal distribution region in high-dimensional space. Moreover, minimum decision function, which indicates distance between...

Gemological Institute of America (GIA) has developed a diamond color measurement instrument that can provide accurate and reproducible results. The uses uniform illumination by daylight-approximating light source; observations from high-resolution color-camera with nearly zero-distortion bi-telecentric lens, image processing to calculate parameters diamonds. Experiments show the results also identify subtle differences in diamonds high sensitivity. experimental setup prototype method for...

In video surveillance fields, the effectiveness of 3D terrain description method is base improving monitoring effect, selecting positions cameras to be deployed reasonably and scientifically.In this paper, a new local land surface roughness representation introduced, then combined with features camera forms measurement merging neighbouring regions.Experiments show results proposed simplification method, also proved.For real data used in experiments, number grids reduced sharply.The...

FastMap algorithm can embed the nodes of a given edge-weighted undirected graph into Euclidean space in near-linear time. And distance between two this approximates length shortest path them graph. We present new variant called with SPFA (FMS). FMS be implemented an easy way and its preprocessing time is less than original version. Experimental results demonstrate that, although has same runtime performance former Dijkstra's (FMD) when generated heuristics are used A*, preprocesses much faster.

Machine Reading Comprehension (MRC) research concerns how to endow machines with the ability understand given passages and answer questions, which is a challenging problem in field of natural language processing. To solve Chinese MRC task efficiently, this paper proposes an Improved Extraction-based method Answer Re-ranking (IERC-AR), consisting candidate extraction module re-ranking module. The uses improved pre-training model, RoBERTa-WWM, generate precise word representations, can...