Transhumance-Aware Through Corridors: Towards Addressing Farmer-Herder Conflicts in the Sahel

This paper presents an innovative approach based on reinforcement learning to maintain livestock in transhumance corridors in the Sahel. The Q-Learning algorithm is used to model movements in four scenarios. The first involves a shepherd exploring the corridor alone, followed by the animals. The second relies on the shepherd's knowledge to guide his herd. The third involves autonomous animals controlled by smart collars. The fourth involves simultaneous learning by the shepherd and the animals. Movements are modeled virtually, with behavior influenced by interactions between shepherd and animals as a function of distance from corridor boundaries. This indirect model, via collars, enables the animals to stay within the set limits. This approach paves the way for evaluations in real-life conditions. It could help reduce conflicts between farmers and breeders, exacerbated by demographic pressure and the scarcity of resources.