Cloud computing optimizes service quality and resource efficiency via centralized hardware and computational resources. However, the predominantly centralized deployment and operation of cloud data centers increase the physical distance to end-users, leading to degraded service quality. Edge computing addresses this by offloading data processing and analysis tasks directly to devices at the network edge, reducing reliance on backhaul transmission and thus offering a more responsive solution for latency-sensitive applications. Nevertheless, ensuring that applications meet predefined Service Level Agreement (SLA) in resource-constrained edge environments remains challenging. To tackle these issues, this paper investigates elastic scaling strategies in cloud-edge collaborative settings. We propose an attention-enhanced bidirectional LSTM model (A-Bi-LSTM) for microservice workload prediction, and design an adaptive elastic scaling system named XScale. This system incorporates a fall-back scaling mechanism when predictions are unreliable and introduces a proactive load forwarding strategy to enhance overall edge node performance. Experimental results show that, compared to existing elastic scaling methods, XScale reduces SLA violations by 82.3%, increases average resource utilization by 17.4%, decreases average response time by 21.1%, and improves overall edge node performance by 36.3%.

The high-quality development of the iron and steel industry not only promotes the industrialization process but also provides a solid material foundation for infrastructure construction, manufacturing upgrading, and technological innovation. The iron and steel industry must transform and upgrade due to environmental pressures and green transformation, resource constraints and cost pressure, market demand and overcapacity, and various other challenges. Hence, an integrated intelligent management and control platform for steel based on CISDigital is developed based on the needs of the iron and steel industry. On one hand, a smart factory system architecture based on CISDigital is established from the aspects of operation, production, and decision-making, which lays the data and architecture foundation for the application of artificial intelligence technology in the iron and steel industry. On the other hand, leveraging cutting-edge artificial intelligence technologies and integrating knowledge, a series of products have been innovatively developed, which realize the monitoring, prediction, optimization, and decision-making in the entire steel smelting process. Numbers industry-leading intelligent manufacturing projects have been successfully implemented, ranging from individual processes to the whole plant, and the iron and steel industry to the non-ferrous metal industry, which bring hundreds of billions of economic benefits to customers.

ABSTRACT Microservice technology, as a flexible application architecture, has gained wide popularity in the field of Internet of Things (IoT). IoT applications are highly sensitive to latency, making it crucial to place microservices on appropriate edge servers in an edge computing environment. Failure to do so can significantly impact service quality and degrade user experience, posing a major challenge. Addressing the aforementioned issues, this paper proposes a multiobjective service deployment strategy for IoT devices based on reinforcement learning. The goal is to minimize service access delay for IoT devices and reduce the average energy consumption of edge servers in the context of mobile edge computing. To achieve this, we first establish a stochastic optimization model using the Markov decision process (MDP) framework to handle service deployment and resource allocation dynamically. This model captures key characteristics such as heterogeneity in edge server capabilities, dynamic geographic information of IoT devices, and uncertainty in microservice requests. To overcome challenges related to dimensionality, slow convergence, and the exploration–exploitation tradeoff in traditional reinforcement learning algorithms, we introduce deep reinforcement learning into the optimization of microservice deployment. Specifically, we propose the use of deep deterministic policy gradient (DDPG) to obtain a near‐optimal service deployment strategy without manual instructions. DDPG leverages the depth of the network to guide policy gradients and generate solutions that effectively balance exploration and exploitation. To evaluate the proposed approach, we implement the DPG‐MSP (DDPG‐based MicroService Placement) algorithm using real datasets and synthetic data. Comparative analysis with existing microservice deployment algorithms demonstrates the superiority of DDPG‐MSP in terms of performance, robustness, and scalability.

Optimizing multi-instance service composition and dynamic request routing has become a critical challenge in cloud-edge collaborative systems. Existing solutions struggle with effectively balancing performance, cost, and bandwidth constraints in dynamic and resource-constrained environments. In this work, we address these challenges by proposing the Removed Minimum Cost Flow (RMCF) algorithm, aiming to minimize average response time while considering constraints such as budget and bandwidth, making it well-suited for time-sensitive services in a cloud-edge collaborative service provision system. Simulations were conducted using real-world data from China Telecom’s base stations in Shanghai, and experiments in various scenarios were considered, including time-sensitive services and general application services, with key performance indicators such as time utility, cost-utility, request completion rate, and timeout rate. The experimental results demonstrate that RMCF achieves lower response times, superior performance, and improved cost-effectiveness compared to other baseline algorithms.

In mobile edge computing (MEC) environment, effective microservices deployment significantly reduces vendor costs and minimizes application latency. However, existing literatures overlook the impact of dynamic characteristics such as the frequency of user requests and geographical location, and lack in-depth consideration of the types of microservices and their interaction frequencies. To address these issues, we propose TP-MDU, a novel two-stage deployment framework for microservices. This framework is designed to learn users’ dynamic behaviors and introduces, for the first time, a minimal deployment unit. Initially, TP-MDU generates minimal deployment units online, tailored to the types of microservices and their interaction frequencies. In the initial deployment phase, aiming for load balancing, it employs a simulated annealing algorithm to achieve a superior deployment plan. During the optimization scheduling phase, it utilizes reinforcement learning algorithms and introduces dynamic information and new optimization objectives. Previous deployment plans serve as the initial state for policy learning, thus facilitating more optimal deployment decisions. This paper evaluates the performance of TP-MDU using a real dataset from Australia’s EUA and some related synthetic data. The experimental results indicate that TP-MDU outperforms other representative algorithms in performance.

The latency of cloud-hosted composite applications increases due to extended transmission time from the centralized cloud to end-users, compromising service quality. Typical AI application scenarios like autonomous driving and smart cities demand low network latency. Edge computing addresses this by enabling data collection and analysis in nearby edge data centers, reducing user response time. However, in a complex edge-cloud computing environment, finding the optimal scaling scheme dynamically is crucial due to varying user response times and dynamic scaling costs near edge data centers. This paper proposes a predictive scaling method to adjust microservice container number based on user request fluctuations. Our prediction algorithm, a two-way GRU with an attention mechanism named A-Bi-GRU, aims to minimize scaling jitter. To achieve this, we introduce the concept of an observation window and employ a multi-objective optimization algorithm based on improved NSGA-II, named DP-GA, for microservice scaling across different locations within each window. The solution aims to minimize the average user response time and scaling costs, enabling intelligent dynamic scaling based on location awareness. Experimental results indicate that the proposed A-Bi-GRU forecasting algorithm achieves approximately a 30% improvement in prediction accuracy over traditional linear models such as LR and SVM, and about a 5–10% improvement compared to conventional recurrent neural networks like RNN and LSTM. Furthermore, the proposed DP-GA multi-objective optimization algorithm reduces average response time by roughly 80% and scaling cost by approximately 50%.

Abstract Container virtualization technology represented by Docker has been widely used in the industry due to its advantages of lightweight, fast deployment, and easy portability. It can bring convenience to system deployment, operation, and maintenance. This article considers the implementation of AI‐based IoT applications based on Docker and container technology in a cloud‐edge collaborative environment. Three typical data‐intensive applications, preprocessing, training, and inference in machine learning, are deployed using Docker containers. For these three types of tasks, this article proposes a container‐based data‐intensive application scheduling framework. Using an attribute‐based data‐driven method, a new scheduling approach considering multiple weighting factors is proposed, which is called DICS‐OPT. The cloud nodes and edge nodes are scored and sorted uniformly. The container is scheduled to the node with the highest score, and then the task data is transmitted to the node to execute the task. NSGA‐III‐based genetic algorithm has been proposed to search for the optimal weighting factors. This article introduces the implementation of the prototype system and builds a cloud‐edge collaborative testbed consisting of Raspberry Pis and PCs. The performance evaluation results indicates that the proposed scheduling approach outperforms existing approaches. Compared with existing approaches, DICS‐OPT improves the average edge resource utilization by 10.05% to 69.04%, and saves the cloud resource cost by 16.02% to 36.68%.

With the large-scale commercial use of 5G technology, the era of Mobile Edge Computing with the Internet of Everything as the core is opening. Various computing resources are deployed to the edge of the network near the mobile smart terminal, forming a mobile edge environment for numerous application scenarios. Under this environment, the mobile edge network needs to use the path selection method to obtain one or more service data transmission paths and seamlessly migrates the service data to the most appropriate edge server, to ensure the continuity of edge services and reduce the resource occupation of the mobile edge network. Therefore, this paper proposes a method of Cost-awareServiceMigration Path Selection based on DeepReinforcement Learning (SMART), aiming to jointly optimize communication costs and communication delays under the premise of meeting service requirements. This method transforms the service migration path selection problem in the mobile edge environment into a bi-objective optimization problem under dual constraints, i.e., to find low-latency, low-cost and high-quality service migration paths while satisfying the constraints of computing power resources and transmission time of mobile smart terminals. Then, a DQN is used to construct the corresponding Markov chain decision model according to the problem scenario to find the optimal path for edge service migration. The proposed method learns to select the optimal edge service migration path through the interaction with the environment without obtaining a large amount of historical edge service migration path information in advance. The experimental results onShanghai (Beijing) Telecom mobile communication base station dataset and Shanghai (Beijing) taxi trajectory dataset show that the proposed method can efficiently select low-latency, low-cost, high-quality edge service migration paths in mobile edge environment when vehicles move continuously.It outperforms six typical edge service migration path selection methods, i.e.,Q-learning, A-Star, PLP, PLP/F, PLP/P, and Dijkstra, by at least 15% in all evaluation metrics except computational time.

In traditional non-orthogonal multiple access (NOMA) systems, the secondary user often encounters decoding challenges caused by a lower signal-to-noise ratio (SNR) as it is assigned less power to prevent the primary user from decoding errors. One way to address this issue is by employing semantic communication, which is known for its robustness in low SNR conditions. In this paper, multi-user semantic communication is investigated in a traditional-semantic hybrid NOMA (TSH-NOMA) system, enabling simultaneous transmissions from both the bit user and the semantic user at the same frequency. However, the compatibility between continuous semantic signals and discrete bit signals remains an open problem. Quantization of semantic features has been attempted to address this problem, but it often results in noticeable performance degradation. To tackle this issue, a novel digital semantic constellation design that allows the encoder to generate a semantic constellation similar to typical digital modulation is proposed. This enables the semantic user’s signal to be readily transmitted over the digital channel without affecting the traditional bit user. Simulation results demonstrate that the proposed method allows for the transmission of the semantic user’s signal on the digital channel with negligible performance degradation. Furthermore, the secondary user in the proposed TSH-NOMA system exhibits considerable performance improvement over its counterpart in traditional NOMA, particularly at low-to-medium SNR, without compromising the performance of the primary user.

With the popularity of cloud native and DevOps, container technology is widely used and combined with microservices. The deployment of container-based microservices in distributed cloud-edge infrastructure requires suitable strategies to ensure the quality of service for users. However, the existing container orchestration tools cannot flexibly select the best deployment location according to the user’s cost budget, and are insufficient in personalized deployment solutions. From the perspective of application providers, this paper considers the location distribution of users, application dependencies, and server price differences, and proposes a genetic algorithm-based Internet-of-Things (IoT) application deployment strategy for personalized cost budgets. The application deployment problem is defined as an optimization problem that minimizes user service latency under cost constraints. This problem is an NP-hard problem, and genetic algorithm is introduced to solve the optimization problem effectively and improve the deployment efficiency. The proposed algorithm is compared with four baseline algorithms, Time-Greedy, Cost-Greedy, Random and PSO, using real datasets and some synthetic datasets. The results show that the proposed algorithm outperforms other competing baseline algorithms.

The capability of humanoid robots to generate facial expressions is crucial for enhancing interactivity and emotional resonance in human-robot interaction. However, humanoid robots vary in mechanics, manufacturing, and ap-pearance. The lack of consistent processing techniques and the complexity of generating facial expressions pose significant challenges in the field. To acquire solutions with high confidence, it is necessary to enable robots to explore the solution space automatically based on performance feedback. To this end, we designed a physical robot with a human-like appearance and developed a general framework for automatic expression generation using the MAP-Elites algorithm. The main advan-tage of our framework is that it does not only generate facial expressions automatically but can also be customized according to user preferences. The experimental results demonstrate that our framework can efficiently generate realistic facial expressions without hard coding or prior knowledge of the robot kinematics. Moreover, it can guide the solution-generation process in accordance with user preferences, which is desirable in many real-world applications.

We propose the Unified Transformer for Facial Reaction GeneratioN (UniFaRN) framework for facial reaction prediction in dyadic interactions. Given the video and audio of one side, the task is to generate facial reactions of the other side. The challenge of the task lies in the fusion of multi-modal inputs and balancing appropriateness and diversity. We adopt the Transformer architecture to tackle the challenge by leveraging its flexibility of handling multi-modal data and ability to control the generation process. By successfully capturing the correlations between multi-modal inputs and outputs with unified layers and balancing the performance with sampling methods, we have won first place in the REACT2023 challenge.

Semantic communication systems have been proposed for efficient text transmissions in recent years. However, most existing methods mainly focus on word-level recovery, which fails to reflect the semantic fidelity of the model. By leveraging the sentence-level semantic information, a semantic communication system architecture is proposed within the framework of deep learning based joint source-channel coding. Then, in order to evaluate the semantic fidelity of the system, a new metric of Semantic Similarity is proposed, which is more sensitive to semantic differences as compared to existing evaluation metrics, e.g., bilingual evaluation understudy. For guaranteeing the consistency between model training and performance evaluation, the proposed new metric is then incorporated into the objective function, based on which end-to-end performance optimization is performed. Extensive simulation results on European Parliament dataset demonstrate the effectiveness and necessity of the proposed method. Compared with the state-of-the-art, e.g., DeepSC, an improvement of up to 10% is achieved in terms of Semantic Similarity. Furthermore, significant performance gains are achieved by the proposed method in both regimes of low and medium signal-to-noise ratio, as compared to traditional separate source-channel coding methods.

In Mobile Edge Computing (MEC) environment, service deployment for IoT application is a key issue that needs to be solved. Considering the knowledge of mobile users’ service requests and edge server’s processing capacity, the problem of microservice deployment in MEC environment is modelled as a non-linear optimization problem. An adaptive dynamic deployment optimization method called Adapt-SD has been proposed, which is based on Adam and weighted round-robin scheduling algorithm to solve this microservice deployment problem. In Adapt-SD, considering the hardware resource-constrained MEC environment, different numbers of microservice instances are deployed on different edge servers, and then microservice instances are invoked to achieve the minimum resource consumption cost while meeting user’s service access delay constraints. At the same time, Adapt-SD also ensures the work balance of edge servers. In this paper, real datasets from EUA in Australia and some synthetic datasets are utilized to measure the performance of Adapt-SD, which is compared with the existing microservice deployment algorithms. Experimental results show that Adapt-SD is superior to other representative deployment algorithms.

The rapid growth in the number and diversity of Web service, coupled with the myriad of similar Web service in functionality, makes it challenging to find most suitable Web service for users to accelerate and accomplish Mashup development. Therefore, this article proposes a Web service recommendation method via integrating heterogeneous graph attention network representation and FiBiNET (Feature Importance and Bilinear feature Interaction NETwork) score prediction. In this method, first, a heterogeneous information service network is constructed by using composite service information, atomic service information, and their respective attribute information. Second, the meta-paths are defined according to different semantic information and service similarity matrixes are built by using commuting matrix and meta-path-based similarity measurement technology. A two-layer attention model is designed to calculate the node level attention and meta-path-level attention of the services respectively, and generate the feature representation of Web service. Third, for the Web services in the service cluster, combining their feature representations with multi-dimensional QoS attributes, the FiBiNET is exploited to dynamically learn the importance of features and complex feature interactions, and predict the score of Web services. Finally, the experiments are performed on the real Web service dataset. The experimental results show that the proposed method is better than the other nine methods in terms of accuracy, recall, F1, and AUC, and achieves better classification and recommendation quality.

Mobile edge computing (MEC) has become an extremely hot topic in recent years. Mobile edge cloud relies on storage and computing resources on network edge to provide users with delay-sensitive services. However, the transmission delay among microservices and the load of the servers tend to increase due to improper service placement and unreasonable resource allocation under MEC. In this paper, an edge service placement strategy based on an improved fast non-domination sorted genetic algorithm is proposed. First, a microservice placement optimization model is built with the goal of minimizing the average transmission delay and the load balance degree. Then, a genetic algorithm-based microservice placement approach called GA-MSP using improved NSGA-II is studied under the premise that a single service instance is deployed only on one container. The experiments show that the proposed GA-MSP approach is able to achieve low delay and load balance effectively, and ultimately deploy services based on the resulting sets after convergence, which outperforms several other existing representative methods.

Container virtualization technology represented by Docker has been widely used in the industry due to its advantages of lightweight, fast deployment, and easy portability. It can bring convenience to system deployment, operation and maintenance. This paper considers the implementation of AI-based IoT applications based on Docker and container technology in a cloud-edge collaborative environment. Three typical data-intensive applications, preprocessing, training, and inference in machine learning, are deployed using Docker containers. For these three types of tasks, this paper proposes a container-based data-intensive application scheduling framework. Using an attribute-based data-driven method, a new scheduling approach considering multiple weighting factors is proposed, which is called DICS-OPT. The cloud nodes and edge nodes are scored and sorted uniformly. The container is scheduled to the node with the highest score, and then the task data is transmitted to the node to execute the task. This paper introduces the implementation of the prototype system and builds a cloud-edge collaborative test platform consisting of Raspberry Psi and PCs. The performance evaluation results indicates that the proposed scheduling approach outperforms existing scheduling approaches in improving the utilization rate of edge resources and saving cost of cloud resources.

In this paper, we propose a hybrid storage framework WukaStore, which offers a scalable, configurable and reliable big data storage service by integrating stable storage (such as Cloud storage or durable storage utilities) and volatile storage (such as idle storage harnessed from desktop PCs over the Internet). By configuring different storage strategies, WukaStore delivers cost-effective storage service to satisfy users’ requirements. We present trace-driven simulations to study the impact of different strategies and how to ensure high availability and durability in WukaStore. We present the prototype implementation of WukaStore based on BitDew, an open source middleware for Big Data management. We conduct performance evaluations to validate the effectiveness of WukaStore, through the deployment of WukaStore in the French Grid'5000 experimental platform. In particular, we also present a case study where WukaStore is deployed in a hybrid environment taking advantage of storage provided by Amazon S3, Dropbox and a local Desktop Grid at the same time. Our evaluation results show that WukaStore provides users with an alternative method to store their big data on hybrid volunteered cloud and desktop systems and to decrease the cost of data storage, while providing great data access performance, scalability as well as reliability.

Fog computing is a decentralized computing infrastructure in which data, compute, storage and applications are located somewhere between the data source and the cloud. It usually adopts convenient and flexible distributed services, which can realize low-cost and real-time data analysis and intelligent control. Efficient communication and fog/edge collaboration have become popular research issues. In this paper, the offloading problem of dependent tasks in fog/edge collaborative environment is studied. Dependent task is modeled as a directed acyclic graph (DAG), and the scenario that fog nodes are configured with heterogeneous multi-core servers is considered. According to task dependencies and energy consumption requirements, all subtasks executed on different edge devices are prioritized, and the Priority and Dependency-based DAG Tasks Offloading Algorithm (PDAGTO) is proposed. Simulation results have shown that, compared with the existing work, the proposed algorithm can effectively reduce the average delay and the total energy consumption of during the procedure of task offloading.

Web service composition has become a prevalent software development method that enables developing powerful Mashups by effectively combining Web services with different functions. However, as the number of Web services increases, it becomes challenging for developers to select appropriate services to develop Web applications that satisfy functional requirements. In order to recommend Web services considering user's preferences, a composition pattern-aware Web service recommendation method called EWACP-DeepFM is proposed, which combines the composition patterns between Web services and Mashups and the co-occurrence and popularity of Web services. By constructing a multi-dimensional feature matrix, which is further trained by the depth factorisation machine (DeepFM) model to learn potential link relationships between Web services and Mashup applications, and recommend Top-N best services for the target Mashup application. Experiments performed using the real datasets from ProgrammableWeb show that the proposed method outperforms others with better recommendation effectiveness.

Summary This paper introduces HybridMR, a novel model for the execution of MapReduce (MR) computation on hybrid computing environment. Using this model, high performance cloud resources and heterogeneous desktop personal computers (PCs) in Internet or Intranet can be integrated to form a hybrid computing environment. Thanks to HybridMR, the computation and storage capability of large scale desktop PCs can be fully utilized to process large scale datasets. HybridMR relies on two innovative solutions to enable such large scale data‐intensive computation. The first one is HybridDFS, which is a hybrid distributed file system. HybridDFS features reliable distributed storage that alleviates the volatility of desktop PCs, thanks to fault tolerance and file replication mechanism. The second innovation is a new node priority‐based fair scheduling (NPBFS) algorithm has been developed in HybridMR to achieve both data storage balance and job assignment balance by assigning each node a priority through quantifying CPU speed, memory size, and input and output capacity. In this paper, we describe the HybridMR, HybridDFS, and NPBFS. We report on performance evaluation results, which show that the proposed HybridMR not only achieves reliable MR computation, reduces task response time, and improves the performance of MR, but also reduces the computation cost and achieves a greener computing mode. Copyright © 2015 John Wiley & Sons, Ltd.

Cloud SLAs are contractually binding agreements between cloud service providers and cloud consumers. For cloud service providers, it is essential to prevent SLA violations as much as possible to enhance customer satisfaction and avoid penalty payments. Therefore, it is desirable for providers to predict possible violations before they happen. We propose an approach for predicting SLA violations, which uses measured datasets (QoS of used services) as input for a prediction model. As a feature of cloud service, we consider response-time to predict violations of SLA. The prediction model is based on Naive Bayesian Classifier, and trained using historical SLA datasets. We present the basics of our prediction approach, and also determine the most effective combinations of features for prediction, and briefly validate our approach, using a detailed real SLA datasets of cloud services. Experiments result show that the Bayesian method achieves higher accuracy compared with other prediction methods.

With data intensive applications it can be interesting to resort to a distributed storage to reach scalability and avoid data-intensive problems. Storing data permanently on computing nodes can be an interesting approach especially with the frequent use and the large volume of this data. Moreover, processing large data is a computing intensive task which encourages parallel execution. Nevertheless, data placement on computing nodes should be optimal to reach load balancing. In this work, we investigate scheduling heuristics towards the optimization of data distribution on the computing nodes. Motivated by its capacity to control perfectly the common operations associated with data management, we use BitDew: a desktop grid middleware designed for large scale data management. With BitDew, we build a Data-Driven Master/Worker Platform to carry out the distribution of Magick, the OCR application based on Dynamic Time Warping (DTW) algorithm. We evaluate the benefit of the implementation of studied scheduling heuristics to achieve load balancing with both homogeneous and heterogeneous environment. We present experimental results which demonstrate the efficiency of our aproach.

There exists numerous Grid middleware to develop and execute programs on the computational Grid, but they still require intensive work from their users. BitDew is made to facilitate the usage of large scale Grid with dynamic, heterogeneous, volatile and highly distributed computing resources for applications that require a huge amount of data processing. Data-intensive applications form an important class of applications for the e-Science community which require secure and coordinated access to large datasets, wide-area transfers and broad distribution of TeraBytes of data while keeping track of multiple data replicas. In genetic biology, gene sequences comparison and analysis are the most basic routines. With the considerable increase of sequences to analyze, we need more and more computing power as well as efficient solution to manage data. In this work, we investigate the advantages of using a new Desktop Grid middleware BitDew, designed for large scale data management.Our contribution is two-fold: firstly, we introduce a data-driven Master/Slave programming model and we present an implementation of BLAST over BitDew following this model, secondly, we present extensive experimental and simulation results which demonstrate the effectiveness and scalability of our approach. We evaluate the benefit of multi-protocol data distribution to achieve remarkable speedups, we report on the ability to cope with highly volatile environment with relative performance degradation, we show the benefit of data replication in Grid with heterogeneous resource performance and we evaluate the combination of data fault tolerance and data replication when computing on volatile resources.

Abstract Interactive generation of falling motions for virtual character with realistic responses to unexpected push, hit or collision with the environment is interesting work to many applications, such as computer games, film production, and virtual training environments. In this paper, we propose a new method to simulate protective behaviors in response to the ways a human may fall to the ground as well as incorporate the reactive motions into motion capture animation. It is based on simulated trajectory prediction and biomechanics inspired adjustment. According to the external perturbations, our system predicts a motion trajectory and uses it to select a desired transition‐to sequence. At the same time, physically generated falling motions will fill in the gap between the two‐motion capture sequences before and after the transition. Utilizing a parallel simulation, our method is able to predict a character's motion trajectory real‐time under dynamics, which ensures that the character moves towards the target sequence and makes the character's behavior more life‐like. Our controller is designed to generate physically plausible motion following an upcoming motion with adjustment from biomechanics rules, which is key to avoid an unconscious look for a character during the transition. Based on a relatively small motion database, our system is effective in generating various interactive falling behaviors. Copyright © 2006 John Wiley & Sons, Ltd.