In this paper, we present noise reduction for a hose-shaped rescue robot. The robot is used for searching for disaster victims by capturing their voice with its microphone array. However, the ego noise generated by its vibration motors makes it difficult to distinguish human voices. To solve this problem, we propose a noise reduction method using a blind source separation technique based on independent vector analysis (IVA) and noise cancellation. Our method consists of two steps: (1) estimating a speech signal and an ego noise signal from observed multichannel signals using the IVA-based blind source separation technique, and (2) applying noise cancellation to the estimated speech signal using the estimated ego noise signal as a noise reference. The experimental evaluations show that this approach is effective for suppressing the ego noise.