Mimicking animal behavior: Implementing reinforcement learning strategies on a wireless robot
Francisco Bastos Pereira · Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT) · 2025
Animals constantly modify their behavior based on the outcomes of previous experiences; a process we commonly call learning. This work seeks to replicate certain biological learning processes in a mobile robotic platform, as the study of this processes in both biological and artificial systems has long been a subject of interest in neuroscience and robotics. In particular, the research focuses on operant conditioning, a learning mechanism in which an agent’s behavior is shaped by associating actions with a reinforcement, or punishment stimuli. In the mammalian brain, the basal ganglia is believed to play a key role in this process, implementing a kind of reinforcement learning framework that guides behavior through dopaminergic signals from the midbrain. To explore these principles, the study will utilize a wireless mobile robot developed at the Champalimaud Foundation. The robot will be trained to chase or avoid “objects” according to whether these have previously provided positive or negative feedback. These "objects" consist of images projected onto the ground by a ceiling-mounted projector and detected by the robot using a wireless camera. The ultimate objective is for the robot to learn associations between visual inputs and behavioral responses, ensuring it moves towards rewarding stimuli while avoiding negative ones. Throughout the project, we implement Reinforcement Learning techniques, specifically Temporal-Difference Learning. The programming of the robot as well as the experimental platform will is done by using Python and Bonsai, a reactive-based visual programming language, capable of naturally handling asynchronous streams of data.