We investigate the response of groups of zebrafish, a model social organism, to a free-swimming robotic fish. The robot has a body and tail section and moves forward by beating the tail. Steering control is achieved by offsetting the beating tail with respect to the body. The color pattern and shape of the robot are informed by visual cues known to be preferred by zebrafish. A realtime multi-target tracking algorithm uses position and velocity estimates to autonomously maneuver the robot in circular trajectories. Observables of collective behavior are computed from the fish trajectory data to measure cohesiveness, polarization, and speed of the zebrafish group in two different experimental conditions. We show that while fish are significantly less polarized in the presence of the robot with an accompanying change in average speed, there is no significant change in the degree of cohesion.