Innovative sensor technologies and perception methods enable Physical AI systems to understand their environment reliably under real-world conditions. The integration of heterogeneous, multimodal sensor data with learning-based semantic representations and uncertainty quantification provides the basis for physically grounded state estimation.