INTELLIGENT SOFTWARE FOR CONTROLLING ROBOTIC SYSTEMS USING VIRTUAL MODELING

Аннотация

The integration of artificial intelligence (AI) into mobile robotic systems is redefining the landscape of autonomous navigation, decision-making, and task execution. Traditional rule-based systems, while effective in structured environments, often fail in dynamic or unknown terrains. This research investigates the application of AI techniques – namely fuzzy logic and reinforcement learning – in developing an intelligent navigation system for mobile robots within a virtual simulation environment. Using the CoppeliaSim Edu platform, a three-wheeled differential-drive mobile robot was modeled, equipped with various sensors (infrared, ultrasonic, compass), and placed in a randomized maze-like environment. The robot’s goal was to autonomously navigate the environment, avoid obstacles, and reach predefined targets efficiently. Initially, a fuzzy rule-based controller was implemented to provide baseline navigation logic. Subsequently, a Q-learning reinforcement learning agent was introduced to optimize path planning through experience-based learning. The combination of these techniques led to a marked improvement in navigation performance, as evidenced by reduced completion times and increased obstacle avoidance efficiency. Graphical data and tabular analysis support the effectiveness of AI-based control versus traditional logic. The research confirms that virtual environments, coupled with intelligent algorithms, can serve as powerful tools for developing and testing advanced robotic control systems before deployment in physical hardware. This approach is cost-effective, safe, and scalable for educational, research, and industrial applications. Keywords: AI Navigation, Mobile Robots, Simulation, CoppeliaSim, Reinforcement Learning, Sensor Fusion, Path Planning The field of robotics is rapidly shifting toward fully autonomous and intelligent systems capable of operating in unstructured, dynamic environments. The growing demand for robots in logistics, service sectors, smart manufacturing, and autonomous vehicles has underscored the necessity for adaptive navigation systems that go beyond hardcoded logic. Classical methods – based on deterministic algorithms – lack the flexibility and responsiveness needed for real-time decision-making in uncertain scenarios. Artificial intelligence (AI) offers a transformative solution to these https://scientific-jl.com 5 4 59-son_4-to’plam_Aprel-2025 ISSN:3030- 3 6 1 3 T A D Q I Q O T L A R jahon ilmiy – metodik jurnali limitations. By enabling machines to learn from experience, interpret sensor data contextually, and adapt to environmental changes, AI techniques like fuzzy logic and reinforcement learning have become essential tools in modern robotics. Fuzzy logic enables robots to handle ambiguity in sensor data, making it suitable for approximate reasoning in complex environments. On the other hand, reinforcement learning (RL) empowers a robot to learn optimal policies through trial-and-error interactions, maximizing cumulative rewards without explicit programming. This research aims to combine both methods in a simulated robotic system. The simulation environment provides a cost-effective, safe, and scalable alternative to physical prototyping, which is especially important in the early stages of AI algorithm development. CoppeliaSim Edu is selected for its versatility, sensor simulation capabilities, and compatibility with control algorithms. The core objectives of the study are: To model a mobile robot with realistic motion and sensing in a virtual maze environment. To implement fuzzy logic for initial control and obstacle avoidance. To train a Q-learning-based RL agent to improve navigation through autonomous learning. To evaluate the performance of AI-driven control through metrics such as time- to-goal, path length, and collision rate. This study contributes to the broader field of intelligent mechatronics by offering a blueprint for integrating AI navigation logic into simulation platforms, laying the groundwork for eventual deployment on real-world robotic systems. Simulation Setup Robot type: Three-wheeled differential drive Sensors: 3 IR, 1 ultrasonic, 1 compass Control algorithm: Fuzzy logic (initial rules), then deep Q-learning (AI improvement) Environment: Randomized maze with moving and static obstacles Navigation was initially handled using fuzzy logic rules such as: If left IR sensor is near and front sensor is far⇒Turn right Later, reinforcement learning updated actions based on reward signals: 𝑄(𝑠,𝑎) ← 𝑄(𝑠,𝑎) + 𝛼[ 𝑟 + 𝛾 ∗ 𝑚𝑎𝑥 𝑄(𝑠′,𝑎) − 𝑄(𝑠,𝑎)] (1) 𝑎 Where: s - current state, a - action, α - learning rate, γ - discount factor, r - reward The robot was tested over 50 simulation trials. The graph below shows the average time taken to reach the goal with and without AI learning: Trial Range Without AI (sec) With AI (sec) 1-10 34.2 29.5 https://scientific-jl.com 5 5 59-son_4-to’plam_Aprel-2025 ISSN:3030- 3 6 1 3 T A D Q I Q O T L A R jahon ilmiy – metodik jurnali