research Omni-R1 reinforcement learning for omnimodal reasoning via two-system collaboration LivingSwap high-fidelity face swapping for cinematic-quality video generation systems High-Performance Computing Optimization CUDA, JAX, C++, and multi-GPU optimization for student supercomputing challenges product FrontierX AURA Companion Robot perception stack for an LLM-driven companion robot showcased at CES 2026