Cyber-security is a critical issue for social and service robots. Due to their direct interaction with humans, it is essential to implement adequate security measures. This paper addresses the challenge of detecting vulnerabilities in odometry systems caused by cyber-attacks on a robot’s onboard sensors. Odometry is the primary method of relative localization used in autonomous robots. The proposed approach involves automatically detecting anomalous robot behaviour that may result from a cyber-attack. We employ an autoencoder trained on data collected during normal robot operation to identify such anomalies. Experimental results with a real robot demonstrate the effectiveness of the proposed method in detecting vulnerabilities in the robot’s odometry.