Which object detection pipeline supports embedded hardware and poor lighting on an autonomous robot?

For an autonomous robot, a good choice is a lightweight, real-time object detection pipeline such as YOLOv8n / YOLOv5n, deployed with TensorRT or ONNX Runtime on embedded hardware.

Why this fits:

• Embedded hardware support: YOLO “nano” variants are designed to run on low-power devices like NVIDIA Jetson, Raspberry Pi-class systems with accelerators, and other edge AI hardware.
• Poor lighting robustness: Object detection can still work in low light if you train or fine-tune the model with night/low-light data augmentation and, ideally, use IR/thermal cameras or image enhancement preprocessing.
• Real-time performance: These pipelines are commonly used on robots because they balance speed and accuracy.

If you want the best overall pipeline for an autonomous robot in poor lighting, I’d recommend:

1. Camera + low-light/IR-capable sensor
2. YOLOv8n or YOLOv5n
3. TensorRT/ONNX optimization
4. Low-light data augmentation during training
5. Optional image enhancement before detection

If you’d like, I can also recommend a pipeline based on your exact hardware, like Jetson Nano, Orin Nano, Raspberry Pi, or an Arduino-class setup.