49. Edge AI and hardware acceleration

Run perception and control models under tight power, memory, and latency limits using quantization, pruning, accelerators, and efficient runtimes. Benchmark the full sensor-to-action path on target hardware.