5 on stationary obstacles placed in the vehicle’s path.
Driving models can recognize hazards yet fail to brake
Experiments on Alpamayo-1.5 separate hazard perception from action control and show that inference-time guidance can reconnect them without retraining.
Big Tech
Matteo Cercola · Katie Z Luo · Simone Formentin · Marco Pavone
Politecnico di Milano · Stanford University · NVIDIA
Research Digest··3 min read
Cercola and colleagues investigate why a reasoning driving model sometimes identifies an obstacle but still plans through it.
§