Active localization and navigation
This research aims to develop a navigation and path-planning algorithm that integrates visual SLAM with reinforcement learning for a ground-aerial multi-vehicle system. Autonomous vehicles have become essential for extreme environments such as search and rescue, disaster relief, and infrastructure inspection, where human presence is limited. The combination of unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UAVs) enhances mission efficiency by leveraging the complementary strengths of both systems—UGVs provide endurance and payload capacity, while UAVs offer aerial surveillance and rapid maneuverability.
Localization and mapping are crucial for swarm robotic systems to achieve coordinated autonomy in dynamic environments. Unlike single-robot SLAM, swarm robotics requires decentralized or cooperative approaches to share positional and environmental data among multiple agents efficiently. Techniques such as multi-agent SLAM, topological mapping, and probabilistic localization enable robust navigation while minimizing computational and communication overhead. Sensor fusion, incorporating LiDAR, IMU, and vision-based methods, enhances accuracy in GPS-denied environments. Focus on implementing localization and mapping without using GPS for swarm system which includes UGVs and UAUs. Advanced strategies like collaborative loop closure, distributed pose graph optimization, and reinforcement learning improve adaptability and resilience in uncertain conditions. The integration of machine learning and neural networks further refines localization accuracy, making swarm robotic systems more efficient for applications in search-and-rescue, exploration, and industrial automation.
What are sensors? Sensors are devices that measure and record various properties. In robotics, we use sensors to gather information about a robot’s internal parameters and its environment. Essentially, sensors function as the robot’s “eyes” and “ears”. Sensor types Proprioceptive – Sensors that measure internal parameters of the system.(gyroscopes, accelerometers) Exteroceptive – sensors that measure environmental parameters. (Cameras, microphones, LiDAR) Passive – sensors that measure the energy entering the sensor. (cameras, microphones) Active – sensors that measure a parameter by emitting energy to the environment. (sonar, radar) Challenges
IMU Sensor An IMU is an electronic device that integrates an accelerometer, gyroscope, and magnetometer to measure an object’s acceleration, angular velocity, and direction of the geomagnetic field. The Inertial Navigation System (INS) is a dead-reckoning navigation system commonly used for mobile robot navigation. The trajectory of an indoor AMR is determined by integrating acceleration and angular velocity based on the initial position, direction, and velocity. IMUs have advantages such as high output update rate and immunity to external interference, but they have limitations when applied to mobile robots. These disadvantages include the calculation process relying on initial conditions and the need for double integration of measurements. To overcome these limitations, IMUs are often used in fusion with other sensors and as auxiliary measurements in other navigation methods.
Ultrasonic Sensor Ultrasonic sensors, operating at frequencies above 20 kHz, are essential for distance measurement and navigation in mobile robots. They use two primary methods:
LiDAR LiDAR (Light Detection and Ranging) uses laser beams to measure the position and distance of objects by analyzing reflected signals. Operating on the time-of-flight (ToF) principle, LiDAR provides high accuracy, speed, and range compared to conventional radar. In mobile robot navigation, 2D LiDAR is commonly used for indoor applications, while 3D LiDAR is preferred in autonomous driving for detailed environmental mapping. Key advancements include SLAM (Simultaneous Localization and Mapping) techniques such as Gmapping, Hector SLAM, Karto SLAM, and Cartographer, which improve mapping accuracy and efficiency. Recent developments integrate neural networks, intensity-based methods, and multi-sensor fusion to enhance localization and obstacle detection. LiDAR’s limitations, such as errors with glass or mirror surfaces, can be mitigated by fusing ultrasonic sensors.
Simultaneous Localization and Mapping (SLAM) is very important topic in robotics and automation. When operating in unknown environments SLAM provide privilege to while operating also update the localization and map the environment in parallel. SLAM give more efficiency and the autonomy of the robotic system. The algorithms are varied with the sensors which used for the system and the hardware and computational power used on robots. In localization and mapping reducing the uncertainties is a major thing it can be achieved by increasing the accuracy of sensor data but in real world we can’t handle everything as operator needed like drift, wind, slipping and noises from environment etc. So, in case of increasing the accuracy and confidentiality of a localization and mapping system SLAM algorithm act a critical role. Methods like feature extraction, feature matching, extended Kalman Filter, gaussian filters, sign distance function and probability distribution models used to overcome that challenges and increase the efficiency of the autonomous systems. SLAM algorithms have evolved significantly, driven by the increasing demands for accuracy, robustness, and real-time performance in robotics and automation. Feature extraction, probabilistic modeling, and sensor fusion are key trends enabling robust SLAM systems. Modern techniques, such as those leveraging visual-inertial data (e.g., VINS-Mono, ORB-SLAM2) and collaborative approaches, demonstrate the power of integrating diverse sensor data and multi-agent collaboration. Direct methods like DSO and LSD-SLAM have gained prominence for their ability to work effectively in dynamic or low-texture environments. Emerging trends also include integrating LiDAR and visual data, enhancing SLAM performance in low-light or large-scale settings. The field is shifting towards lightweight, scalable solutions capable of handling heterogeneous sensors, real-time constraints, and privacy concerns while maintaining high accuracy in challenging scenarios.
What is localization? Localization refers to the process by which a robot determines its location and understands its surroundings. This is similar to how humans identify their position using environmental cues gathered through their senses, such as sight and sound. Humans can interpret what they perceive and make decisions based on that information. In contrast, robots receive these inputs as signals, which humans can program the robots to analyze and act upon. Typically, humans are quite adept at accurately identifying their surroundings. However, robots can encounter errors in determining their location and environment. When a robot identifies its position, it may be in a known space—where it already possesses a map—or in an unknown space—where it constructs a map as it navigates. The robot should have a belief about its current position. This belief can be about a single position or the robot can identify a set of possible positions of it. These two methods are called • Single hypothesis belief • Multiple hypothesis belief Each methods have their pros and cons Single hypothesis belief Pros: No position ambiguity Cons: robot motion often induces uncertainty due to affective and sensory noise. Multiple hypotheses belief This allows the robot to track not just a single possible position but a set of possible positions. Pros: the robot can reason about reaching a particular goal and consider the future trajectory of its belief state. Cons: the calculations require more computational power. some of the robot’s possible positions imply a motion trajectory that is inconsistent with some of its other possible positions
Map representations