Autonomous cleaning robots (ACRs), also known as cleaning robots or intelligent vacuum cleaners, are a typical application of mobile intelligent robots entering the home. They combine the core technologies of robots and vacuum cleaners, enabling them to easily and automatically clean indoor environments such as homes, hotels, and office buildings without human intervention. They integrate numerous disciplines including mechanics, electronics, sensor technology, computer technology, control technology, robotics, and artificial intelligence. As a pioneer in the practical development of intelligent mobile robots, research on vacuum robots began in the 1980s, and it remains one of the most challenging and popular research topics in the field of home appliances. Currently, both domestic and international research and development of vacuum robots have yielded certain results, and some finished products are on the market. Although autonomous vacuum robots have been commercialized, their performance still needs further improvement. 1. Behavior-Based Programming: The inclusive architecture mentioned in Rodney Brooks' 1986 paper indicates the formal origin of behavior-based programming methods. Inclusive Structure: In the course of evolution, humans never lose their relatively primitive and rudimentary brain engineering, while higher functions are added on top of this. Therefore, each person's brain retains remnants of rudimentary consciousness similar to that of reptiles. Similarly, a behavior-based approach requires designing a series of simple behaviors for the robot (behaviors are algorithms that control the execution process through sensory information). These behaviors coordinate and cooperate to produce the desired overall robot behavior. The system's behavior is not completely deterministic but contains many random elements. The execution process is not entirely stable, but the overall behavior of the system is very stable. Behavior-based robots will connect sensor information directly to the execution process as much as possible. They are highly reflexive: as soon as the robot makes a judgment about the relevant environment, it immediately takes action. It acts immediately upon receiving information. 2. Behavior Design Behaviors are divided into two types: servo behavior and ballistic behavior. Servo behavior uses a feedback control loop as its control unit. Ballistic behavior operates according to a pre-defined pattern from beginning to end. The overall planning process of ballistic behavior is closely related to the implementation code. Any minor errors in the environment or behavior initialization process (such as misoperation caused by noise or false signals) during execution can cause problems for the robot, leading to complete failure. Servo behavior has good noise immunity and strong fault tolerance to other minor faults during operation. The behavior in this paper uses a finite state machine (FSM) to make it easier to understand the system's working process, thus making it easier to write the system implementation code. 2.1 Cruise Behavior Cruise behavior is the most basic and most frequently used behavior of robots. This behavior causes the two drive motors to output the same speed, and the robot moves forward in an approximately straight line until other behaviors are triggered. When other behaviors finish running, it will return to cruise behavior. 2.2 Wall-Walking Behavior Wall-walking behavior can help the robot search for paths among obstacles. Especially in multi-room environments, performing a short wall-walking behavior near a door will give the robot a better chance to enter other rooms. Therefore, when the robot encounters an obstacle, it needs to perform a short wall-walking behavior at intervals. As shown in Figure 1: 2.3 Homeward Navigation The homeward navigation mechanism, combined with the berth sensor, allows the robot to return to the charging station when its battery is low, ensuring task completion. The performance of the two infrared beacon receivers for the robot's homeward navigation cannot be completely identical. When the robot determines that it is directly facing/facing the light source by comparing sensor outputs, its actual orientation is biased towards the light source. The robot moves along a spiral towards the beacon's position. The robot rotates while moving forward, with a rotation angle ω = k(L - R), where k is the gain parameter; L and R are the received signal strengths of the infrared receivers. When the robot's battery is low, it may not be in the charging room and therefore cannot detect the infrared beacon signal. In this case, the wall-walking behavior should be triggered to move the robot to a room where the beacon signal can be detected before triggering the homeward navigation. As shown in Figure 2. 2.4 Escape Behavior The escape behavior protects the robot from danger and ensures the successful completion of the task. The robot's rotation angle is an important parameter used to balance the system's environmental adaptability. If the value is large, the robot can leave walls or other large obstacles very cleanly and efficiently; however, the robot loses its ability to find narrow paths, severely limiting its ability to navigate autonomously in complex environments. If the angle is small, the robot will be able to navigate through cluttered environments more easily, but it will need to repeat the process multiple times to successfully avoid walls. Therefore, a random rotation angle should be chosen. As shown in Figure 3. 2.5 Anti-stall behavior The output torque of a permanent magnet DC motor is proportional to the current. When voltage is applied across the motor and the motor is not rotating, the torque and current reach their maximum values. If the robot collides with an obstacle and there is a large friction force between the drive wheel and the ground, the drive motor will stop. Therefore, if the motor has the maximum current and the current has lasted for a relatively long time, it indicates that the robot has collided with an object in the environment. The stall detection sensor can only determine that a collision has occurred when the motor has been in a high current state (above a certain threshold) for a period of time (motor startup will generate an instantaneous current peak). 2.6 Anti-static behavior The virtual static detection sensor can be implemented simply through software. The output information of each sensor may change continuously when the robot is moving, but once it stops moving, the output information of all sensors will remain unchanged. 2.7 System Structure The robot's function and operation mode determine its structure. The system structure diagram is shown in Figure 4. 2.8 Differential Drive The differential drive chassis controls the overall movement of the robot by controlling the difference in motion between the two drive wheels. No matter how complex the motion is, it can be decomposed into translational motion and stationary rotational motion. Figure 5 shows the differential drive model, which describes the relationship between the speed of the two drive wheels and the robot's radius of curvature, where the radius of curvature is rL = VLW / (VR - VL). When the rotational speeds of the two drive wheels are exactly the same, the value of the radius rL will tend to infinity. At this time, the process of the robot moving in a straight line can be understood as the robot rotating along a circle with an infinite radius. When the speed of the left wheel is 0, rL equals 0, and the robot will rotate around the left wheel in place. At this time, vL = v, Vr = wW + v. When the speeds of the left and right wheels are the same but opposite in sign, the robot will rotate around its center position in place (rL ​​= w / 2). A differential drive robot can rotate around any point on the line connecting the axes of the two drive wheels (including points outside the robot body). A negative radius indicates that the robot travels counterclockwise along the arc; a positive radius indicates that it travels clockwise along the arc. 3. Arbitrator Design When only one action is triggered at any given time, the system can operate relatively smoothly. However, when multiple actions are triggered simultaneously, and each action requires the robot to perform different operations, the robot needs to use an arbitration mechanism to properly handle this relationship. Here, a fixed priority arbitration is used. Each action is uniquely assigned a priority value, and when a conflict occurs, the action with the higher priority is executed. After issuing a control request, an action needs to know whether it has been approved by the arbitrator. The arbitrator has a certain output; each action is assigned a unique identifier (ID: Identifier). The arbitrator outputs the identifier of the action that wins the arbitration. Each action can determine whether it has control over the arbitration resource by comparing its own identifier with the arbitrator's output. During robot operation, many uncontrollable factors exist in the environment, and the actual operation of the robot differs greatly from the expected behavior. Sometimes sensors fail completely, and sometimes false alarms and missed detections frequently occur during information detection (a missed alarm means that a sensor cannot detect information that should be detected in the environment; a false alarm means that the information detected by the sensor is not present in the environment or is incorrect). Although performance will be affected to some extent in the event of loss of important information or deterioration of motion control commands, the robot program should still strive to perform at its best, rather than completely paralyzing. The ability of a system to continue operating at a reduced level in the event of a subsystem error or failure is called graceful degradation. The robot designed in this paper has a complete graceful degradation function (see Figure 6). In the event of collision sensor failure, anti-blocking and anti-stationary behaviors will ensure that the robot continues to complete the task. 4. Robot Simulation and Analysis The proposed behavior-based vacuum cleaner robot design scheme is simulated. Figure 7 shows the simulation interface of the robot cleaning task in a simulated room. The robot adopts a random coverage mode and does not know its specific location, so it will inevitably revisit areas it has already visited. As the robot operates, the growth rate of the covered area decreases, and the area coverage rate can be approximately expressed as: Coverage rate = (1 - et/a), where f is a time variable and a is a time constant. This indicates the time spent by the robot to perform deterministic coverage without visiting old areas. Experimental simulations show that the design scheme presented in this paper fully meets the requirements. Although the random coverage method lacks the low repeatability of deterministic coverage, it avoids the cost, complexity, and system fragility issues associated with deterministic coverage. 5. Conclusion Behavior-based robot design methods do not use expensive single-type sensors to achieve unattainable accuracy and reliability. Instead, they comprehensively utilize multiple sensor systems with relatively lower reliability, leveraging the complementary advantages of these systems to enhance the robot's robustness. The authors, combining traditional behavior-based and sensor technologies, selected the most effective behavior combinations to construct the entire system, significantly improving its efficiency compared to traditional random coverage.