With the CUDA toolkit and TensorRT inference runtime, autonomous vehicles running on the NVIDIA DRIVE AGX platform can achieve the critical latency requirements for autonomous driving.