Peyman Moghadam

NEO achieves visually coherent and geometrically consistent scene edits that significantly enhance robotic manipulation capabilities, outperforming state-of-the-art methods.

TOLiD achieves superior transfer learning in LiDAR representation by effectively bridging the modality and architecture gaps between vision and 3D data.

SSD transforms radar place recognition by leveraging spatial alignment to overcome the limitations of sparse sensor data, leading to unprecedented accuracy in challenging conditions.

Depth-aware distillation boosts visual place recognition in forests, revealing that geometric cues can dramatically enhance model robustness against appearance variations.

Current vision models falter in natural environments, with the WildCross benchmark revealing critical gaps in depth estimation capabilities.

Residual connections, a staple of modern transformers, actively *harm* quantization robustness by driving activations away from Gaussianity.