Invent and implement new architectures combining perception, visual/language understanding, and control, under tight compute and latency budgets. Work with Vision-Language Agents (VLAs), Vision Transformers (ViTs), and diffusion-based planners.

Push forward the AI component of a new software robotics stack. Invent and prototype new architectures for vision, language, multimodal integration, and control. Deploy models on real robots with quick iteration cycles.

Design and build experimental setups across real-world robotic systems and simulation environments. Calibrate hardware, optimize data pipelines, and ensure reproducibility and scale.