Research
Core Research Areas
Pioneering the convergence of 6G Wireless Communications, Artificial Intelligence, and Robotics to build the intelligent infrastructure of the future.
Intent-Driven AI-RAN &
Wireless Foundation Models
We design AI-native RAN intelligence that translates high-level human intent into real-time network actions. An Intent LMM interprets operator inputs and orchestrates control apps across the O-RAN RIC stack.
Simultaneously, a Wireless Foundation Model fuses digital-twin data with field CSI measurements to predict channel states, enabling proactive resource allocation on GPU-accelerated PHY/MAC platforms.
- Intent LMM: Translates natural language to structured network policies.
- Proactive Control: Resources allocated efficiently before demand peaks.
- GPU-Accelerated: High-reliability services via NVIDIA Sionna and Aerial frameworks.
Distributed Learning &
Privacy-Preserving Computation
We study distributed learning for battery-limited wireless edge systems. Devices update lightweight models using local data and share privacy-preserving parameters with a central server, while the edge runs capable Large Multimodal Models (LMMs).
Our framework performs channel-aware offloading, adapting data transmission based on time-varying wireless conditions to meet strict Quality of Service (QoS) targets for both accuracy and latency.
- Split Computing: Joint optimization of Device and Edge computation capabilities.
- Channel-Aware: Adaptive task offloading based on real-time wireless channel states.
- Privacy-First: Secure model updates without sharing raw local data.
Connected Collaborative Robotics:
Fleet-Level Navigation
We implement a two-level intelligence loop for connected robotics. Robots execute fast on-device control for immediate safety and actuation, while an Edge LMM aggregates multi-robot telemetry to form a global traffic map.
The Edge LMM reasons over fleet-wide context to provide coordinated high-level guidance. Unlike fully on-device systems limited by local views, our Edge-Holistic approach prevents congestion and scales reliably as the fleet grows.
- Two-Level Loop: Combines fast local control with intelligent global planning.
- Edge LMM: Resolves fleet-wide congestion and routing conflicts dynamically.
- Scalability: Efficient and reliable coordination for large-scale robot fleets.
