Applied Behavioral Analysis (ABA) techniques are widely used and accepted by the Autism research community as an effective Autism therapy method. ABA techniques have been recently introduced to use with Socially Assistive robots (SAR) to deliver Autism therapy. Nonetheless, little research has been published to investigate the use of robot-based ABA in teaching socio-emotional skills for children diagnosed with Autism Spectrum Disorder (ASD). This paper presents the development of an ABA-based autonomous therapy system delivered by a humanoid robot, the Zeno R-50. Specifically, an intervention methodology with a prompt and an ABA reinforcement protocol to target skills associated with facial and situational emotion recognition and understanding are presented. Eleven children diagnosed with Autism were recruited to participate in the pre-pilot study screening. Initial results are investigated to discover the children’s preferred reinforcers and also to find ways of improving the therapy system before proceeding through full study groups. Results demonstrate the successful detection of reinforcers and show there is correlation between reinforcer preference and age. Results on two children who have completed interventions are presented and improvements to the protocol are discussed to contribute to the understanding of SAR in teaching socio-emotional skills to children diagnosed with ASD.