Robot Imitation Learning: Trajectories Without Encoders https://sudonull.com/robot-imitation-learning-trajectories-without-encoders
Teach a servo-driven manipulator to repeat movements through a neural network and potentiometers. Detailed breakdown of data collection, MLP training, and error compensation. For middle/senior developers.
H.264 vs H.265 in IP cameras: load on server and archive https://sudonull.com/h-264-vs-h-265-in-ip-cameras-load-on-server-and-archive
We analyze why identical codecs in cameras produce different bitrates and load on VMS. Technical factors, encoder settings, and testing methodology.
Legal RAG ARLC 2026: 17 iterations to 0.791 https://sudonull.com/legal-rag-arlc-2026-17-iterations-to-0-791
Breakdown of RAG pipeline on legal challenge: hybrid search, RRF, cross-encoder reranking, fast-paths. From 0.034 to 0.791 and drop on scale. Code and metrics for developers.
SDXS-1B: compact diffusion for GPU https://sudonull.com/sdxs-1b-compact-diffusion-for-gpu
Breaking down the SDXS-1B architecture: UNet 1.6B, Qwen3.5, asymmetric VAE with 2x upscaler. Metrics, training, code. Master image generation on your own GPUs — download the model.