Robots are increasingly used alongside Skinner boxes to train animals in operant conditioning tasks. Similarly, animals are being employed in artificial intelligence research to train various algorithms. However, both types of experiments rely on unidirectional learning, where one partner-the animal or the robot-acts as the teacher and the other as the student. Here, we present a novel animal-robot interaction paradigm that enables bidirectional, or mutual, learning between a Wistar rat and a robot. The two agents interacted with each other to achieve specific goals, dynamically adjusting their actions based on the positive (rewarding) or negative (punishing) signals provided by their partner. The paradigm was tested in silico with two artificial reinforcement learning agents and in vivo with different rat-robot pairs. In the virtual trials, both agents were able to adapt their behavior toward reward maximization, achieving mutual learning. The in vivo experiments revealed that rats rapidly acquired the behaviors necessary to receive the reward and exhibited passive avoidance learning for negative signals when the robot displayed a steep learning curve. The developed paradigm can be used in various animal-machine interactions to test the efficacy of different learning rules and reinforcement schedules.