How robust is today’s AI vision when the world gets messy?

The Mouse vs AI: Robust Visual Foraging Competition, held as part of the NeurIPS Competition Track, turns that question into a direct benchmark. Participants train artificial agents to perform the same visually guided foraging task as mice in a virtual-reality environment, then test how well those agents handle visual perturbations such as fog and other out-of-distribution changes.

The competition asks whether AI agents can match, or even outperform, biological vision under degraded viewing conditions. It also goes one step further: in addition to scoring behavioral robustness, the challenge includes a neural-alignment track that tests whether an agent’s internal visual representations predict cortical activity recorded from mouse visual areas.

The goal is to bring together researchers in reinforcement learning, robust computer vision, systems neuroscience, and biologically inspired AI around a shared task: building agents that do not just perform well in clean training environments, but remain reliable when visual conditions change.

Final submission deadline: November 15, 2025.

Learn more on the competition website: https://robustforaging.github.io