Efficient object search in unknown, cluttered environments remains a critical challenge in robotics, particularly under occlusions and partial observability. This thesis presents HOPE- OS (Heuristic Occlusion-aware Planning for Efficient Object Search), a novel heuristic-based Next Best View (NBV) planning algorithm that combines gradient-based Region of Interest (ROI) maps with occlusion-aware Field of View (FoV) simulation. The method guides the robot to informative viewpoints that maximize visibility while minimizing unnecessary motion. HOPE-OS operates without prior environmental knowledge, supports multiple ROIs and target objects, is robot-agnostic, and achieves improved computational efficiency over existing approaches. Extensive evaluations were conducted in 2D and 3D simulated environments, including a realistic mock-up of CERN’s Large Hadron Collider (LHC), as well as real-world tests using a 9-degree-of-freedom UR10 robot. Notably, we introduce a novel benchmarking protocol comparing robot decisions to human strategies, demonstrating that HOPE-OS achieves human-level search efficiency. The contributions of this work include (1) a real-time, heuristic NBV planning algorithm aware of occlusions, (2) a gradient-based ROI clustering approach for viewpoint sampling, and (3) a human-robot evaluation framework. This research advances the state of the art in robotic object searching, providing a versatile and efficient framework with potential applications ranging from industrial maintenance to search-and-rescue operations.
Efficient object search in unknown, cluttered environments remains a critical challenge in robotics, particularly under occlusions and partial observability. This thesis presents HOPE- OS (Heuristic Occlusion-aware Planning for Efficient Object Search), a novel heuristic-based Next Best View (NBV) planning algorithm that combines gradient-based Region of Interest (ROI) maps with occlusion-aware Field of View (FoV) simulation. The method guides the robot to informative viewpoints that maximize visibility while minimizing unnecessary motion. HOPE-OS operates without prior environmental knowledge, supports multiple ROIs and target objects, is robot-agnostic, and achieves improved computational efficiency over existing approaches. Extensive evaluations were conducted in 2D and 3D simulated environments, including a realistic mock-up of CERN’s Large Hadron Collider (LHC), as well as real-world tests using a 9-degree-of-freedom UR10 robot. Notably, we introduce a novel benchmarking protocol comparing robot decisions to human strategies, demonstrating that HOPE-OS achieves human-level search efficiency. The contributions of this work include (1) a real-time, heuristic NBV planning algorithm aware of occlusions, (2) a gradient-based ROI clustering approach for viewpoint sampling, and (3) a human-robot evaluation framework. This research advances the state of the art in robotic object searching, providing a versatile and efficient framework with potential applications ranging from industrial maintenance to search-and-rescue operations. Read More
Related posts:
Enhancing Event-Based Vision for Logistics: Data Processing, Prediction Models Analisys and 3D Dataset Creation
Optimizing Maintenance of Energy Supply Systems in City Logistics with Heuristics and Reinforcement Learning
Innovative Metaheuristic Optimization Approach with a Bi-Triad for Rehabilitation Exoskeletons
Navigation of mobile robots using neural networks and genetic algorithms
Nature-Inspired Metaheuristic Optimization for Control Tuning of Complex Systems
Real-Time Object Detection for Autonomous Solar Farm Inspection via UAVs