Few-Shot Beam Selection for Embodied AI via In-Context Learning and Conformal Prediction

Millimeter-wave (mmWave) communications are essential for supporting massive multimodal sensory data exchange in Embodied Artificial Intelligence (EAI) systems. However, fast and reliable beam selection for mmWave communication remains a critical bottleneck due to the prohibitive overhead associated with conventional beam training. Although location-aided beam prediction offers a promising alternative, its performance is often limited by localization errors and the need for extensive training data. To address these challenges, we propose a novel framework termed In-Context Learning driven Cross-Validation Conformal Prediction (ICL+CVCP) for beam prediction. The proposed framework utilizes the training-free adaptation capability of ICL, enabling effective operation with only a small number of demonstration samples and eliminating the need for large-scale data collection and model fine-tuning. In addition, the CVCP module constructs compact, distribution-free candidate beam sets that provably contain the optimal beam with a user-defined probability. Experimental results confirm that ICL+CVCP consistently satisfies the desired empirical coverage while significantly reducing pilot overhead and computational latency, thereby meeting the stringent reliability and efficiency requirements of EAI applications.