Frontiers in Embodied AI (ICCV 2025 WDFM-AD)
Contents
Jamie Shotton, Chief Scientist, Wayve
End-to-End Architecture
Challenges of Embodied AI
- driving data is easy to collect.
Generalization of End-to-End Autonomous Driving
from UK to USA, driving on the left side to drive on the right side
Commercial Process
Nissan unveils next-generation ProPILOT
NVIDIA Investment
Language Meets Driving (16:20)
- Common driving sense
Explainability
LINGO-1
https://wayve.ai/thinking/lingo-natural-language-autonomous-driving/
LINGO-2
https://wayve.ai/thinking/lingo-2-driving-with-language/
Future Directions
Spatial Intelligence
extension of VGGT
Expanding the Value of Data
GAIA-2
- generate temporal and spatially consistent multi-view images.
- Pre-collision scenarios
- Emergency braking scenarios
Safety-critical scene editing
Surround from single view
Surround RGB from depth/lidar
Synthetic out-of-distribution states
Learning in Simulation
Embodied AI Vision for Success
