Ergodic Imitation for Adaptive Exploration around Demonstrations

2026-05-13

AI Summary

This work addresses the challenge of imitation learning failure in robotic deployment due to environmental changes or imperfect perception and control. To this end, the authors propose an adaptive ergodic imitation framework that, for the first time, integrates demonstration data into a retrieval-based receding horizon optimization architecture. By retrieving relevant demonstrations to construct a goal distribution, the method dynamically interpolates between trajectory tracking and exploratory behavior during execution. Combining ergodic control, demonstration retrieval, and goal distribution modeling, the approach extends classical ergodic coverage—traditionally used for area sweeping—to adaptive imitation learning. This enables robust task completion under environmental perturbations and significantly enhances policy generalization and robustness.

Abstract

In robotics, a common challenge in imitation learning is the mismatch between training and deployment conditions, caused, for example, by environmental changes or imperfect observation and control. When a robot follows a nominal trajectory under such mismatch, it may become stuck and fail to complete the task. This calls for adaptive online exploration strategies that remain grounded in demonstrations. To this end, we propose an adaptive ergodic imitation approach that constructs a target distribution from the geometry of the retrieved demonstrations and uses it to generate trajectories that adaptively interpolate between tracking and exploration. Our method extends ergodic control beyond its traditional role in area-coverage and search by incorporating demonstrations into a retrieval-based receding-horizon framework for adaptive imitation.

Problem

• imitation learning
• distribution mismatch
• adaptive exploration
• demonstration-based control
• trajectory adaptation

Innovation

• ergodic control
• imitation learning
• adaptive exploration
• receding-horizon planning
• demonstration-based learning