Topic: 3D Scene Reconstruction, Embodied Perception and World Model for Robotics

Driven by the rapid evolution of robotics, computer vision and generative artificial intelligence, robot intelligence has stepped into a new stage centered on 3D spatial cognition and embodied interaction. Robots deployed in diverse real-world environments face prominent challenges including complex unstructured surroundings, multi-modal heterogeneous sensory inputs, cross-domain task migration gaps, and insufficient generalization of environment understanding capabilities, which greatly restrict the reliability and intelligence of robot autonomous operation.

The rapid progress of 3D reconstruction technology, embodied intelligent perception and robotic world models provides solid theoretical support and core technical reserves to break through the above bottlenecks. High-fidelity 3D scene reconstruction can restore geometric, semantic and physical attribute information of real environments; embodied perception endows robots with human-like interactive sensing capabilities through multi-sensor fusion and active exploration; while world models enable robots to predict environmental evolution, simulate physical interaction processes and complete task planning in virtual space. Combined with large-scale embodied simulation training platforms, these technologies effectively solve the pain points of scarce real robot data, high test cost and low safety of physical verification. Such technical frameworks can be generalized to universal application scenarios, covering general daily service robotics, industrial manufacturing, power grid inspection, medical operation and other vertical fields, realizing unified 3D spatial perception and predictive reasoning for multi-industry robot systems.

This Special Issue aims to collect cutting-edge theoretical breakthroughs, algorithm innovations and engineering implementations around robotic 3D scene reconstruction, embodied perception and world models. It builds an open academic exchange platform for robotics researchers, computer vision scholars, industrial R&D engineers and cross-domain engineering practitioners, to exchange frontier research results, discuss existing technical bottlenecks, and jointly promote the theoretical iteration, cross-industry deployment and large-scale industrial landing of robot spatial intelligence technologies.

Topics of interest include, but are not limited to:

● High-precision geometric-semantic fusion 3D scene reconstruction algorithms for robots; 

● Lightweight real-time 3D reconstruction for mobile robots, manipulators and quadruped robots; 

● Multi-modal embodied perception frameworks based on vision, LiDAR, tactile and audio sensors; 

● Active embodied exploration strategies for unknown unstructured environments;

● Construction, optimization and rendering of large-scale embodied robot data training grounds; 

● Robotic world model design, physical reasoning and future state prediction methods; 

● World model driven robot motion planning, manipulation and autonomous decision-making; 

● 3D reconstruction and embodied perception algorithms for general vertical industrial scenarios; 

● Perception system optimization oriented to industrial production lines and power grid inspection robots;

● Cross-domain generalization of 3D perception and world models across service, industrial, energy and medical scenarios; 

● Digital twin, simulation-real cross-domain migration training of robotic perception systems; 

● Edge deployment, lightweight optimization and real-world verification of robot 3D perception modules;

● Safety, robustness and interpretability of embodied perception and world model systems.

Robotic 3D scene reconstruction, embodied perception, robotic world model, embodied simulation training ground, multi-modal sensor fusion, active environment exploration, industrial robot perception, power grid inspection robot, digital twin for robotics, physical simulation & predictive reasoning, unstructured environment cognition, cross-domain robotic perception