Perception, representation, and understanding of the environment for autonomous mobile systems (T2)

Objectives: The fundamental challenge of perception, representation, and understanding of the environment in the context of autonomous mobile systems is to develop robust applications, while simultaneously reducing the number of false positives and false negatives in perception, to provide complete, consistent, and accurate representations, and to deliver a correct understanding of the scene in its full complexity.

Research challenges / Novelty / Innovation
• Semantic perception (2D) – segmentation, detection, classification, tracking
• Geometric perception (3D) – 3D reconstruction, point cloud processing: detection, classification, segmentation, tracking
• Improving the generalization capability of deep learning models
• Multi-sensor and/or temporal fusion of semantics with geometry in a cooperative approach
• Neural reconstruction and scene understanding
• Mapping and localization
• Scene understanding and knowledge graph generation

Research results:
• Demonstrator: Open-source platform for perception, representation, and understanding of the environment for drones

Innovation:
• Original methods for panoptic and video panoptic segmentation, monocular depth estimation with metric scaling, 3D point cloud processing, improving the generalization capability of deep learning models, multi-sensor and/or temporal fusion of semantics with geometry in a cooperative approach, neural reconstruction and scene understanding, knowledge graph generation.