
ECCV 2024
Unified Local-Cloud Decision-Making via Residual Reinforcement Learning
K. Sengupta · Z. Shangguan · S. Bharadwaj · S. Arora · E. Ohn-Bar · R. Mancuso
UniLCD dynamically routes inference between on-device and cloud models, using reinforcement learning to balance latency, energy and safety for real-time embodied navigation.
35% improvement over split-computing and early-exit baselines
- Embodied AI
- Reinforcement Learning
- Edge/Cloud



