Peking University Researchers Achieve 30% Battery Life Improvement for Embodied AI Systems with CREATE Framework

Researchers from Peking University's School of Integrated Circuits and School of EECS have identified a critical challenge in deploying embodied AI systems: the substantial computational demands, particularly for portable, battery-powered devices. Their groundbreaking work introduces CREATE, a novel cross-layer resilience framework that synergistically optimizes energy use and reliability. The CREATE framework achieves remarkable energy efficiency gains: up to 40.6% computational energy savings and a 37.3% reduction in chip-level energy consumption, ultimately extending battery life by up to 30%. This research represents a significant step toward practical, robust, and efficient embodied AI agents capable of operating in real-world environments. Modern embodied AI systems frequently integrate a Large Language Model (LLM)-based planner for high-level task management with a reinforcement learning (RL)-based controller for precise action execution, enabling agents to navigate complex real-world scenarios. The core of CREATE lies in three key innovations: anomaly detection and clearance at the circuit level to suppress large errors induced by timing violations, weight-rotation-enhanced planning at the model level to redistribute LLM activations, and autonomy-adaptive voltage scaling at the application level to dynamically adjust the controller's operating voltage. Extensive experimentation has confirmed that CREATE achieves these substantial energy reductions without compromising the reliability or performance of embodied AI agents in complex tasks. The authors acknowledge limitations including the use of a uniform error model and INT8 quantization, which may not fully capture the complexity of real-world error distributions.