Abstract: With the rapid development of ubiquitous computing technology, embedded intelligent devices, represented by smartphones, PDAs, and portable media centers, are constantly evolving and maturing. These devices typically come equipped with customized embedded operating systems on their respective hardware platforms and have an increasing number of applications. Testing these application software programs presents a significant challenge to software providers and device manufacturers. This paper introduces an agent-based testing method into an automated testing platform. By using a test control agent, the script's versatility is improved; by using a target device agent, the complexity of embedded devices is shielded; and by using a network environment agent, control over the network environment in which the device operates is achieved. This effectively assists testers in completing the testing of embedded intelligent devices. Keywords: pervasive computing ; automated testing; Agent technology; embedded devices and phone manufacturers. This paper introduces Agent-based testing technology to the automated testing platform. The test control Agent improves the reusability of the testing scripts. The device Agent reduces the complexity introduced by the embedded devices. And the network environment Agent can help control the network environment of the smart devices. These can help the testers to fulfill their tests on the embedded devices more efficiently. Key words: pervasive computing; automated testing; agent technology; embedded device 1 Preface Intelligent software agents are software programs capable of performing specific tasks for users, possessing a certain degree of intelligence, autonomously executing some tasks, and interacting with the environment in an appropriate manner. Agents have characteristics such as autonomy, responsiveness, learning ability, and sociality. This makes them suitable for making timely responses in highly dynamic environments. The development of embedded technology and next-generation mobile communication networks has led to a surge in embedded intelligent devices. Most of these devices have embedded operating system support and run increasingly rich applications. How to test these applications has become a research topic. This paper introduces agent technology into the testing of embedded intelligent devices, using target device agents, test control agents, and network environment agents to simulate and handle the complexity of test equipment, test engineers, and test environments, respectively. Utilizing the characteristics of agents themselves, an effective automated testing method is proposed. 2 Related Research and the Idea of ​​This Paper The autonomy, responsiveness, learning ability, and sociality of agents make them well-suited for handling problems in complex testing systems. There has been some research on applying agent technology to the testing field, which is introduced in detail below. Jeongeun Choi and Byoungju Choi [1] proposed a test tool based on Agent technology. It uses a user interface Agent to handle the interaction with the tester, uses a test case selection Agent to select test cases, and uses a regression test Agent to perform regression testing, thus achieving automated software testing. However, the test case selection technology they proposed can only select the best test cases from a large number of existing test cases and cannot reduce the complexity of writing test cases themselves. Other studies have applied Agent technology to testing in a certain field. Yu Qi, David Hung and Eric Wong [3] proposed a Web application testing method based on Agent technology. They used Agent technology to reduce the complexity of web application testing. Their method is not only applicable to the testing of Web applications, but also suitable for the testing of embedded smart devices. In the actual testing process, a large task for testers is still to write various test cases. To truly improve testing efficiency, it is necessary to improve the universality of test scripts and reduce the changes in test scripts [2] [4]. Common embedded smart device testing tools (such as TestQuest) use image comparison to determine the state of the target device. While this method achieves non-intrusive testing, it has two problems: ● Image capture and transmission consume significant testing resources. ● User interface styles vary greatly across different mobile phones, requiring script maintenance when testing new devices. However, in real-world devices, the system generates sensitive events to incentivize relevant modules to process them when important events occur. Capturing these sensitive events during testing allows for more efficient testing. Furthermore, as long as the operating system is the same, the method of capturing sensitive events remains consistent, and test cases do not change due to interface variations. This paper uses a target device agent to abstract and capture sensitive events during the testing process and sends these events to a test control agent. Upon receiving a sensitive event, the test control agent, based on its knowledge, takes appropriate action (such as exception handling or test rescheduling). This serves as the agent's inference rule, using the agent's intelligence to shield the complexity of the testing process. Furthermore, considering the complex network environment in which the target device is located, we utilize a network environment agent to control the network signal of the target device, thereby achieving control over the network environment in which the device is located. 3 Agent-based Testing System 3.1 Overall Structure The ultimate goal of the testing system is to realize an automated testing platform for mobile digital terminals, capable of establishing a modular and reusable test script library for system software and application layer software on mobile digital terminals, improving the efficiency of various black-box tests, automating the generation of test results, and realizing the playback of test results. [align=center] Figure 1. Overall Architecture of the Testing System[/align] As shown in Figure 1, the testing platform is divided into four layers, including the user interface layer, test control layer, communication layer, and device agent layer. Among them, the process monitoring module, test environment control module, and device agent layer are designed using the agent concept to better support the testing of embedded devices. 3.2 Basic System Execution Flow The testing process is a simulation of the "input-feedback" model when a real user uses a mobile phone. The basic use cases of the system are as follows: First, the user creates a test script and performs basic configuration through a script editor or virtual phone. Then, the process monitoring module schedules the script to start execution. When the script interpreter interprets and executes the script statements, the script interpreter usually sends simulated key presses to the mobile phone through the communication module. Then, the script interpreter enters a waiting state. The target device agent receives the key presses from the communication module, executes the key press actions, captures the screen and returns it to the test control agent. Alternatively, if a sensitive event occurs on the mobile phone (such as an incoming call, SMS, anomaly, etc.), it will also provide feedback to the test control agent. Upon receiving the data, the test control agent will again stimulate the script to run and simultaneously input and store the test results into the test resource library. The script continues to execute the following statements until the execution ends. 3.3 Target Device Agent The target device agent resides in the smart device under test and realizes the modeling of the target device. In this system, it receives control commands from the PC and then performs corresponding operations on the smart device, including simulating keyboard events, capturing the screen and returning it to the PC, and, based on the test objectives configured in the knowledge base, selectively and proactively notifying the test host of necessary status information through reasoning. 3.3.1 Decision Rules Dependent on Device Agent ● System Event Reporting Decision Rule (SERD-RULE): This decision rule determines whether the device agent should report status information to the upper-level system when it receives it. ● Exception Handling Rule (EM-RULE): This decision rule determines the handling method to be taken when the agent detects an exception during testing. 3.3.2 Agent Attributes The target device agent has the following attributes: ● Autonomy The target device agent allows test users to focus on the details of the device without having to pay attention to them. It is a good abstraction of the device's functions and status. Through the target agent, the upper-level testing framework only needs to execute the corresponding virtual operations. The target device agent will automatically map these virtual operations to the actual operations on the mobile phone, and the mobile phone's response and other status can also be automatically reported to the client. This makes the upper-level testing independent of the underlying device's specific operations. ● Intelligence When the target device agent obtains a changed state or information of the system under test, it will send only the information relevant to the current target to the test control agent according to the inference rules. The target device agent can also reasonably handle exceptions that occur during testing according to the exception handling rules. 3.4 Test Control Agent The Test Control Agent models the testing process. It primarily achieves three goals: ● During testing, it captures events of interest to testers. When these sensitive events occur, the system immediately executes predefined actions and provides accurate and detailed logs for the test. ● Through interaction with the script interpreter, it enables asynchronous event notifications and the execution of interactive test cases. ● It handles exceptions during testing, allowing the testing process to self-correct and ensuring that multiple scripts run normally according to a pre-defined schedule. 3.4.1 Decision Rules of the Test Control Agent ● Event-Condition-Action Rule (ECA-RULE) This is the core rule upon which the Test Control Agent operates. It defines the events of interest to a particular test and takes corresponding actions based on comparison conditions. ● Exception Handling Rule (EM-RULE) The Test Control Agent also adheres to the exception handling rule. It detects abnormal events occurring during testing and executes corresponding solutions, ensuring the smooth progress of the test. 3.4.2 Attributes of the Test Control Agent <1> Autonomy: Autonomy allows testers to focus less on the details of the test. Testers only need to select the test target, and the Test Control Agent can automatically generate an ECA rule table based on information in the knowledge base. During script execution, it captures sensitive events and performs corresponding operations according to the ECA rules. <2> Intelligence: The Test Control Agent can automatically generate an ECA rule table based on the user-selected test target, and perform reasoning and take corresponding actions according to the ECA rules. The Test Control Agent can also handle exceptions that occur during the test according to exception handling rules, ensuring the automatic and smooth progress of the test process. 3.5 Network Environment Agent The Network Environment Agent simulates and controls the network environment. To test the radio frequency performance of mobile phones, a base station simulation system is usually used to provide the wireless network environment in which the device is located. A programming interface is used to control the device's environment through software, making it possible to test the target device in complex network environments. 4 System Evaluation To evaluate the effectiveness of this system, we designed a comparative experiment. The experiment was conducted in a real test project at the Telecommunication Technology Labs of the Ministry of Information Industry. We divided our test engineers into two groups. The first group used MobileTest, the system described in this paper, for testing, while the second group used the industry-renowned TestQuest Pro. The testing was based on the digital mobile station testing section of the Chinese telecommunications industry standard. The tests involved performing regression tests on the new smartphone using the regression test cases already established for each system, thus comparing the testing efficiency and script maintainability of the two testing tools. The experiment measured the percentage of automatically executed test cases for different types of tests. As shown in the table, while TestQuest has a higher automation rate in functional testing, MobileTest has better test coverage in stress testing, multi-state testing, multi-task testing, critical testing, and total testing. This is because in the latter four types of testing, test cases need to continuously track the system's state. TestQuest, using image comparison mechanisms, struggles to accomplish this, while MobileTest can handle sensitive events returned by the agent. Furthermore, MobileTest completes its testing tasks in 15 days, which is more efficient than TestQuest. 5. Conclusion This paper proposes a testing method for embedded intelligent devices based on Agent technology. It uses Agent technology to simulate and control the device under test (DUT), the tester, and the network environment during the testing process, effectively shielding the complexity of embedded intelligent devices and improving testing efficiency and script reusability. Future research will further expand the entire testing system to support the entire testing lifecycle. The authors' innovation lies in introducing software Agent technology into the automated testing of embedded intelligent devices. Utilizing the characteristics of Agents, the complexity of the tester, the DUT, and the network environment in embedded intelligent device testing is effectively shielded, improving the efficiency of automated testing. 6 References [1] Jeongeun Choi, Byoundju Choi: Test Agent System Design [A]. Ewha Womans University, Proceedings of 1999 IEEE International Fuzzy System Conference Proceedings, pp326-327, 1999. [2] He Youquan: Software Testing Methods and Techniques [J]. Microcomputer Information, 2000 [3] Yu Qi, David Kung, Eric Wong: An Agent-based Testing Approach for Web Applications [A]. Proceedings of the 29th Annual International Computer Software and Application Conference, pp1-3, 2005. [4] Gu Le, Shi Jiulin. Introduction to Software Testing Technology [M]. Tsinghua University Press. 2004 About the author: Jiang Bo (1981-), male (Han nationality), from Xinxiang, Henan, is a master's student at Beihang University. His main research direction is mobile computing and embedded software testing. Long Xiang (1963-), male (Han nationality), professor, doctoral supervisor, head of the Department of Computer Science, School of Computer Science, Beihang University. Main research areas: embedded systems, computer network security. Gao Xiaopeng (1970-), male (Han nationality), associate professor, School of Computer Science, Beihang University. Main research areas: computer architecture, high-performance software simulators, embedded systems.