Reinforcement Learning Guides Live Fish Schools
This arXiv preprint by Hiroaki Kawashima (submitted March 17, 2026) investigates guiding live fish schools using 2D virtual fish trained with model-free reinforcement learning and evaluated in simulation and real-world trials. Simulations show learned policies succeed even when simulated fish frequently ignore stimuli, while live experiments demonstrate the policy steers schools toward specified target directions and significantly outperforms no-stimulus and heuristic 'stay-at-edge' baselines.
Key Points
- 1Trained 2D virtual fish using model-free reinforcement learning to influence live fish schools.
- 2Demonstrated effective guiding in simulations even when simulated fish often ignored virtual stimuli.
- 3Showed real-world trials steer schools toward targets, outperforming no-stimulus and heuristic baselines.
Scoring Rationale
Novel experimentally validated RL application to collective animal behavior, but limited by preprint status and narrow biological generalizability.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
