Frequently Asked Questions¶
Installation Issues¶
Q: Pixi environment issues¶
A: Try these steps:
Common Issues¶
Q: CARLA Connection Error¶
A: Ensure CARLA server is running, check firewall settings, and verify CARLA_HOME environment variable.
Q: CUDA compatibility issues¶
A: Ensure you have:
- Install NVIDIA drivers (if using NVIDIA GPU)
- Verify CUDA 12.8 installation
- Check GPU memory availability
- PyTorch compiled with correct CUDA version
OpenCDA-MARL Integration¶
Q: How to add new MARL algorithms?¶
A: Check the MARL Framework documentation and API Reference.
OpenCDA Integration¶
Q: Perception module not working¶
A: Check:
activate: truein perception config- PyTorch installation with
--apply_mlflag - Camera/LiDAR sensor configuration
Q: torch.cuda.amp deprecation warnings and boolean flag errors¶
A: If you see warnings about deprecated torch.cuda.amp or errors like Expected 'device_type' of type 'str', got: '<class 'bool'>', this is due to PyTorch API changes.
Solution: Replace the YOLOv5 common.py file with the fixed version:
- Copy
docs/files/common.pyfrom this repository - Replace your cached YOLOv5 file at:
- This fixes both the deprecated API warnings and the boolean flag error
The fixed version uses the new torch.amp.autocast API with proper device type and enabled parameters.
Q: How to create custom scenarios?¶
A: See the YAML Configuration Guide for detailed instructions.
Getting Help¶
Still having issues?
- Check the Original OpenCDA Documentation
- Open an issue on GitHub
- Review the Contributing Guide