CUDA/GPU Compatibility Fix Needed

I have an existing PyTorch-based autonomous driving evaluation pipeline that runs in the CARLA simulator. It's currently failing on newer NVIDIA GPU hardware due to CUDA/driver/dependency compatibility issues.
I need someone to:

Get the pipeline running cleanly on the target GPU environment (debugging CUDA/driver/dependency conflicts)
Run evaluation and confirm standard output metrics are generated correctly
Deliver clean, runnable source code with basic documentation and a demo of it working

Requirements:

Strong PyTorch experience
Experience with CARLA simulator or similar simulation environments
Comfortable debugging CUDA/driver/dependency issues, ideally with newer GPU hardware
Access to a compatible GPU environment (local or cloud)

Please tell me:

Your experience with CARLA and/or autonomous driving models
Your experience with CUDA/GPU compatibility debugging
Estimated timeline
Your quote