Transformer-Based Visual Perception Models for Autonomous Robots

Self-Supervised Learning Models for Autonomous Robotic Navigation

Autonomous robotic navigation has become a key capability for intelligent robots operating in dynamic environments such as warehouses, hospitals, smart cities, agriculture, and autonomous transportation. While supervised learning methods achieve strong navigation performance, they depend on large labeled datasets that are costly and time-consuming to obtain. Self-Supervised Learning (SSL) addresses this limitation by enabling robots to learn robust visual and spatial representations directly from unlabeled sensor data through self-generated learning objectives. This paper reviews recent advances in SSL techniques, including representation learning, contrastive learning, predictive learning, masked image modeling, and multimodal sensor fusion for autonomous navigation. It also examines the integration of data from RGB cameras, LiDAR, IMUs, GPS, depth sensors, and odometry to improve perception, localization, obstacle avoidance, and path planning in unknown environments. Finally, the paper discusses key challenges such as domain adaptation, computational efficiency, safety, and continual learning, highlighting SSL's potential to enable scalable, adaptive, and lifelong autonomous robotic navigation.

Hybrid Deep Learning Frameworks for Robotic Object Recognition

Robotic object recognition is a fundamental capability that enables autonomous robots to interact intelligently with dynamic environments. Traditional vision-based methods, such as SIFT, SURF, HOG, and template matching, perform well under controlled conditions but struggle with variations in lighting, viewpoint, occlusion, and complex backgrounds. Recent advances in deep learning have significantly improved recognition accuracy by automatically learning features from raw image data. However, individual deep learning models often face challenges related to computational cost, inference speed, and limited generalization. This paper proposes a Hybrid Deep Learning Framework that integrates Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), attention mechanisms, and multimodal sensor fusion (RGB, depth, and LiDAR) to enhance recognition accuracy and efficiency. The framework combines local and global feature extraction, adaptive feature fusion, intelligent object recognition, and robotic decision-making for real-time perception and task execution. It supports applications in industrial automation, warehouse logistics, autonomous mobile robots, healthcare, agriculture, and service robotics while improving robustness, scalability, and computational efficiency.

Vision Transformer-Based Autonomous Safe Landing Zone Detection for MAVs

We propose a novel methodology for autonomous landing zone detection in Micro Aerial Vehicles (MAVs) based on Vision Transformers (ViTs). The core contribution of this work lies in demonstrating that transformer-based architectures, originally developed for large-scale vision tasks, can be effectively adapted to safety-critical aerial robotics applications with limited training data. Unlike traditional Convolutional Neural Networks (CNNs), ViTs leverage self-attention mechanisms to model long-range spatial dependencies, enabling a more holistic understanding of scene geometry and surface suitability for landing. We systematically evaluate the proposed approach on aerial RGB images from a public dataset as well as on noisy depth images captured onboard a drone using a lightweight depth camera. Our results show that the ViT-based model consistently outperforms widely used CNN architectures, including ResNet, particularly in low-data regimes where generalization is crucial. Notably, the transformer model maintains strong robustness even when operating on degraded depth inputs. In addition to accuracy improvements, the proposed system achieves real-time performance, with an average inference time of [Formula: see text] ms on legacy GPU hardware. These findings highlight the practical feasibility and effectiveness of Vision Transformers for reliable, efficient MAV landing zone detection.

Self-attention SAC with vision-augmented LiDAR fusion for mapless robot navigation in dynamic environments

Adaptive Vision-Based Navigation in Autonomous Robotic Systems

An intelligent vision-based autonomous robotic framework that integrates deep learning-based object detection with hybrid adaptive navigation for dynamic environments is proposed in this research. The proposed system addresses the challenges of real-time perception and robust navigation in unstructured settings by combining a convolutional neural network (CNN) for object detection with a hybrid control mechanism for motion planning. The CNN, implemented using a state-of-the-art architecture such as YOLO, processes visual input to identify obstacles and target objects, providing critical environmental awareness. Moreover, the hybrid navigation strategy merges reactive obstacle avoidance, achieved through algorithms like the Vector Field Histogram (VFH), with adaptive path planning using Rapidly-exploring Random Trees (RRT) to ensure both immediate collision avoidance and long-term goal convergence. The integration of these components enables the robotic system to dynamically adjust its navigation policy in response to environmental changes, thereby improving robustness and adaptability. The novelty of our approach lies in the seamless fusion of vision-based perception and adaptive control, which enhances the system’s capability to operate in complex, dynamic scenarios. Experimental validation demonstrates the effectiveness of the framework in real-world applications, highlighting its potential for deployment in autonomous vehicles, service robotics, and industrial automation. The proposed method offers a scalable and efficient solution for autonomous systems requiring high levels of situational awareness and adaptive decision-making.

Sensor Fusion and Perception for Autonomous Driving: A Critical Review of Modalities, AI Models, Algorithms, and Industry Configurations

A systematic analysis of the machine learning and deep learning models underpinning vehicle autonomy, spanning classical convolutional neural networks for object detection and semantic segmentation to recurrent and Transformer-based architectures for trajectory prediction and motion planning is presented.