Generative models for grid-based and image-based pathfinding

The use of a policy and a heuristic function for guiding search can be quite effective in adversarial problems, as demonstrated by AlphaGo and its successors, which are based on the PUCT search algorithm. While PUCT can also be used to solve single-agent deterministic problems, it lacks guarantees on its search effort and it can be computationally inefficient in practice. Combining the A* algorithm with a learned heuristic function tends to work better in these domains, but A* and its variants do not use a policy. Moreover, the purpose of using A* is to find solutions of minimum cost, while we seek instead to minimize the search loss (e.g., the number of search steps). LevinTS is guided by a policy and provides guarantees on the number of search steps that relate to the quality of the policy, but it does not make use of a heuristic function. In this work we introduce Policy-guided Heuristic Search (PHS), a novel search algorithm that uses both a heuristic function and a policy and has theoretical guarantees on the search loss that relates to both the quality of the heuristic and of the policy. We show empirically on the sliding-tile puzzle, Sokoban, and a puzzle from the commercial game `The Witness' that PHS enables the rapid learning of both a policy and a heuristic function and compares favorably with A*, Weighted A*, Greedy Best-First Search, LevinTS, and PUCT in terms of number of problems solved and search time in all three domains tested.

To cope with the complexities of flexible manufacturing system FMS scheduling, this paper proposes and evaluates an improved search strategy and its application to FMS scheduling in a Petri net framework. Petri nets can concisely model multiple lot sizes for each job, the strict precedence constraint, multiple kinds of resources and concurrent activities. On the execution of the Petri nets, our algorithm can use both admissible heuristic functions and nonadmissible heuristic functions having the upper supports of the relative errors in A * heuristic search algorithm. In addition, the search scheme can ensure the results found are optimal and invokes quicker termination conditions. To demonstrate it, the scheduling results are derived and evaluated through a simple FMS with multiple lot sizes for each job. The algorithm is also applied to a set of randomly-generated FMSs with such characteristics as multiple resources and alternative routings.

Heuristic search algorithm has been widely used in the fields of geographic information system, intelligent robot path planning, etc. This paper proposes a heuristic search algorithm based on artificial neural network, which is named ANN*. The architecture of the ANN* algorithm is similar to the A* algorithm, but it contains a novel heuristic function of the new design. This novel heuristic function utilizes the regression network, which is a kind of classical artificial neural network. In the ANN* algorithm, the artificial neural network is used to learn the features of massive map data and predict the path search difficulty of the map. In order to guarantee the diversity of the training data, this paper designs a random map generation framework, which is able to generate random grid maps according to the length, width and density. The experiments show that the ANN* algorithm can accomplish the task of path planning in the grid map with high quality. This paper provides a new scheme of exploring the applications of artificial neural network for heuristic search algorithm.

Path planning is one of the important directions in the field of intelligent vehicles research. Traditional path planning algorithms generally use Dijkstra algorithm, Breadth-First-Search (BFS) algorithm and A* algorithm. Dijkstra algorithm is a search-based algorithm, which can search to an optimal path, but the disadvantage is too many expansion nodes, which leads to insufficient search efficiency. BFS algorithm is a heuristic search algorithm, which reduces the disadvantage of too many expansion nodes and improves the search efficiency by heuristic function. A* algorithm is a heuristic search algorithm that combines Dijkstra’s algorithm and BFS algorithm, which has higher search efficiency and can search to an optimal path at the same time, but it is still lacking in the search mode and smoothness of the planned route. This paper first introduces the general path planning algorithm, then introduces and analyzes the A* algorithm, and proposes improvement measures for its shortcomings; finally, the executability and effectiveness of the improved algorithm are tested using simulation, and compared with the traditional A* algorithm, and the results show that the improved A* algorithm has good effect on path planning of intelligent vehicles.

Comparing heuristics has been a major issue from the early days of heuristic search. From the very beginning, measures of the accuracy of heuristic functions were strongly based on the number of nodes generated, and they are often still based on it. In this work, an alternative approach is suggested: to model the accuracy of admissible heuristic functions as the probability of the heuristic hill-climbing search algorithm to find the goal state in a number of steps that equals the heuristic value of an arbitrary node. The resulting method serves to assess on the accuracy of both consistent and inconsistent heuristic functions. Comparisons with different sizes of the sliding-tile puzzle show that the model suggested predicts the accuracy of the heuristic function with a good precision. The resulting procedure is used to derive figures on the accuracy of a large number of heuristics for the 15-Puzzle with different variants of the search algorithm IDA?.

In this paper the number of shortest paths between two points of the square grid using weighted distances is discussed. We use 8-adjacency square grid, that is, the weighted distance depends on the numbers and the weights of the horizontal, vertical and diagonal steps. Two types of neighborhood, and consequently two weights are used. As special cases, the Manhattan distance and chessboard distance, the two well-known and widely used digital distances of the two dimensional digital space occur. Despite our combinatorial result is theoretical, it is closely connected to applications, e.g., in communication networks. The number of shortest paths plays importance in applications of transmitting messages over networks, since they refer somehow to the width of the connection channel between the given points.

This paper considers a problem of planning an attack in robotic football (RoboCup). The problem is reduced to finding a trajectory of the ball from its current position to the opponents goals. Heuristic search algorithm, i.e. A*, is used to find such a trajectory. For this algorithm to be applicable we introduce a discretized model of the environment, i.e. a graph, as well as the core search components: cost function and heuristic function. Both are designed to take into account all the available information of the game state. We extensively evaluate the suggested approach in simulation comparing it to a range of baselines. The result of the conducted evaluation clearly shows the benefit of utilizing heuristic search within the RoboCup context.

In this paper we advance the line of research launched by Knuth which was later improved by Chen for predicting the size of the search tree expanded by heuristic search algorithms such as IDA*. Chen's Stratified Sampling (SS) uses a partition of the nodes in the search tree called type system to guide its sampling. Recent work has shown that SS using type systems based on integer-valued heuristic functions can be quite effective. However, type systems based on real-valued heuristic functions are often too large to be practical. We use the k-means clustering algorithm for creating effective type systems for domains with real-valued heuristics. Orthogonal to the type systems, another contribution of this paper is the introduction of an algorithm called Active SS. SS allocates the same number of samples for each type. Active SS is the application of the idea of active sampling to search trees. Active SS allocates more samples to the types with higher uncertainty. Our empirical results show that (i) SS using clustering-based type systems tends to produce better predictions than competing schemes that do not use a type system, and that (ii) Active SS can produce better predictions than the regular version of SS.

Artificial Intelligence: Search

Existing path planning algorithms are capable of finding physically feasible, shortest, and energy-efficient paths for mobile robots navigating on uneven terrains. However, shortest paths on uneven terrains are often energy inefficient while energy-optimal paths usually take long time to be traversed. Therefore, due to time and energy constraints imposed on mobile robots, these shortest and energy-optimal paths might not be applicable. We propose a multiobjective path planner that can find pareto-optimal solutions in terms of path length and energy consumption. It is based on NAMOA* search algorithm that utilizes a proposed monotone heuristic cost function. The simulation results show that nondominated path options found by the proposed path planner can be very useful in many real-world applications.

An innovative methodology is proposed to identify potential risk factors and possible accident escalation consequences, and to determine the evolution of an accident from cause to consequence, thereby to identify the most probable path and discover key risk factors along the path rapidly. Based on the principle of a directed weighted complex network (DWCN), the bow-tie (BT) model, risk entropy and the improved ant colony optimization (IACO) algorithm are integrated into this methodology. First, the qualitative analysis of risk evolution based on the BT model is carried out. The evolution development based on accident suppression can be divided into two stages: accident precursor stage and accident evolution stage. Then, a new method for mapping BT into DWCN is proposed. Lastly, the shortest path analysis of risk evolution based on the IACO algorithm is carried out, fuzzy set theory (FST) is introduced to calculate the failure probability of risk factors, and risk entropy is used to represent the uncertainty of risk propagation. Thus, the IACO algorithm can be used to calculate the shortest path of risk evolution. The proposed method is applied to oil and gas leakages in the FPSO oil and gas processing system. The results show that it is an effective method to identify the shortest evolution path and the most vulnerable risk factors.

Cultural relics contain unique and rich historical, artistic, economic and other diversified values. However, the imitation and counterfeiting of cultural relics have brought great challenges to the protection and research of cultural relics. In order to improve the problems of strong subjectivity and low precision of the traditional "ophthalmic" identification method, this paper combines spectroscopy with cultural relics to identify the authenticity of pottery and porcelain cultural relics. Under constant temperature and humidity conditions, four kinds of LED with different spectral bands were used as experimental light sources, two Jingdezhen moon flower winter melon porcelain vases with the same appearance and two songxianghu kiln white glazed small clay pots with the same appearance were used as cultural relics samples, and multiple feature points were selected for precision spectral detection. Treat the detection results as multidimensional vectors, calculate the Euclidean distance, Manhattan distance and Chebyshev distance between the vectors, and look for the appropriate threshold to identify the authenticity of the artifact. The results show that: First, the spectral difference between the feature points of cultural relics with the same appearance is significant, and the source of cultural relics can be accurately identified by using the spectral curve; Secondly, this paper calculates the thresholds based on Euclidean distance, Manhattan distance and Chebyshev distance for porcelain and pottery. Among them, the accuracy of identifying porcelain using a threshold based on Euclidean distance is the highest, which is 87.5%; The accuracy of identifying pottery using a threshold based on Manhattan distance was the highest, at 91.67%. By using the spectral curves and thresholds of cultural relics measured by precision spectrometers, this paper generates the spectral identity card of cultural relics. Compared with traditional cultural relics detection, the spectral dimension can better meet the needs of archaeologists for in-depth research on cultural relics.

An important task in signal processing is to estimate the correlation between two signals by distance indices, which directly restricts the validity and reliability of the analysis results. This paper firstly studies the fundamental characteristics of six distance indices, i.e., the Manhattan distance, Euclidean distance, Chebyshev distance, Wasserstein distance, dynamic time warping (DTW), and maximal information coefficient (MIC) distance. Secondly, their confidence intervals are calculated by numerical simulations. The test signal is a standard normal distribution variable plus a deterministic function, i.e., a step (position), a ramp (velocity), a parabola (acceleration) function respectively. A wider range of confidence intervals implied a great uncertainty of a distance index. Simulations show that the Euclidean distance, Chebyshev distance and Wasserstein distance are usually better than the others. Lastly, as an example, using distance values calculated from the moving time window, the optimal lag time between the atmospheric pressure difference from Tianjin to Guangzhou and Guangzhou wind speed is generated, and the future Guangzhou wind speeds can be predicted reliably around the optimal lag time.

One approach to solve detailed routing is net-by-net using a fast A∗-based path search algorithm to handle long connections. Such path search algorithms usually rely on the manhattan distance to implement their heuristic function, providing a poor lower bound since the search is constrained by the global routing guide. This work proposes a new technique to provide a more realistic lower bound in this scenario. We precompute some lower bounds before the path search by applying a modified version of Dijkstra's algorithm on tunnels (sections of the global routing guide). This information is used during the search, reducing the runtime by 58% in average, in comparison to the classic manhattan distance, on ISPD 2018 benchmarks. We also show an improvement of our technique over existing work [10]. Our preprocessing method runtime is negligible, maximizing the benefit of using the improved heuristic function during the path search.

Weighted heuristic search in networks