Robots that learn to evaluate models of collective behavior

2026-04-08

AI Summary

This work proposes an embodied intelligence–based evaluation framework to address the limitations of traditional animal behavior models, which rely on offline static trajectory analysis and fail to capture discrepancies in closed-loop interactions with real organisms. The approach trains a biomimetic robotic fish using reinforcement learning to perform navigation tasks in simulation; the learned policy is then transferred to the physical robot via sim-to-real adaptation and deployed in interactive experiments with live fish. Behavioral fidelity between simulated and real-world interactions is quantitatively assessed using the Wasserstein distance between their respective behavior distributions. This methodology enables, for the first time, a quantitative comparison of behavioral model fidelity grounded in closed-loop interaction, demonstrating that a convolutional neural network–based behavior model significantly outperforms conventional rule-based models in metrics such as target arrival accuracy and exhibits the smallest sim-to-real gap, thereby establishing a new paradigm for evaluating animal behavior models.

Abstract

Understanding and modeling animal behavior is essential for studying collective motion, decision-making, and bio-inspired robotics. Yet, evaluating the accuracy of behavioral models still often relies on offline comparisons to static trajectory statistics. Here we introduce a reinforcement-learning-based framework that uses a biomimetic robotic fish (RoboFish) to evaluate computational models of live fish behavior through closed-loop interaction. We trained policies in simulation using four distinct fish models-a simple constant-follow baseline, two rule-based models, and a biologically grounded convolutional neural network model-and transferred these policies to the real RoboFish setup, where they interacted with live fish. Policies were trained to guide a simulated fish to goal locations, enabling us to quantify how the response of real fish differs from the simulated fish's response. We evaluate the fish models by quantifying the sim-to-real gaps, defined as the Wasserstein distance between simulated and real distributions of behavioral metrics such as goal-reaching performance, inter-individual distances, wall interactions, and alignment. The neural network-based fish model exhibited the smallest gap across goal-reaching performance and most other metrics, indicating higher behavioral fidelity than conventional rule-based models under this benchmark. More importantly, this separation shows that the proposed evaluation can quantitatively distinguish candidate models under matched closed-loop conditions. Our work demonstrates how learning-based robotic experiments can uncover deficiencies in behavioral models and provides a general framework for evaluating animal behavior models through embodied interaction.

Problem

• collective behavior
• behavioral modeling
• sim-to-real gap
• bio-inspired robotics
• model evaluation

Innovation

• reinforcement learning
• biomimetic robotics
• closed-loop evaluation
• behavioral modeling
• sim-to-real transfer