# DropoutDAgger: A Bayesian Approach to Safe Imitation Learning

arXiv: https://arxiv.org/abs/1709.06166 · HTML: https://mlsanity.com/paper/1709.06166/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 18 Sep 2017.

## Built on by (most cited first, 14 of 14)

- [Safe Imitation Learning via Fast Bayesian Reward Inference from Preferences](https://mlsanity.com/paper/2002.09089/) (2020)
- [On the Sample Complexity of Stability Constrained Imitation Learning](https://mlsanity.com/paper/2102.09161/) (2021)
- [Safe end-to-end imitation learning for model predictive control](https://mlsanity.com/paper/1803.10231/) (2018)
- [SAFARI: Safe and Active Robot Imitation Learning with Imagination](https://mlsanity.com/paper/2011.09586/) (2020)
- [Interactive and Hybrid Imitation Learning: Provably Beating Behavior Cloning](https://mlsanity.com/paper/2412.07057/) (2024)
- [DADAgger: Disagreement-Augmented Dataset Aggregation](https://mlsanity.com/paper/2301.01348/) (2023)
- [Agnostic Interactive Imitation Learning: New Theory and Practical Algorithms](https://mlsanity.com/paper/2312.16860/) (2023)
- [FABG : End-to-end Imitation Learning for Embodied Affective Human-Robot Interaction](https://mlsanity.com/paper/2503.01363/) (2025)
- [GrOMP: Grasped Object Manifold Projection for Multimodal Imitation Learning of Manipulation](https://mlsanity.com/paper/2512.03347/) (2025)
- [DISEIL: Demonstration Distillation for Sample-Efficient Imitation Learning](https://mlsanity.com/paper/2609.08123/) (2026)
- [MPC-guided Imitation Learning of Neural Network Policies for the Artificial Pancreas](https://mlsanity.com/paper/2003.01283/) (2020)
- [Imitating careful experts to avoid catastrophic events](https://mlsanity.com/paper/2302.01193/) (2023)
- [ASkDAgger: Active Skill-level Data Aggregation for Interactive Imitation Learning](https://mlsanity.com/paper/2508.05310/) (2025)
- [Stability-Aware Imitation Learning from Model Predictive Control for Autonomous Vehicle Lateral Control: Exact Q-Loss and a Novel Training Procedure](https://mlsanity.com/paper/2609.23506/) (2026)

## Abstract (arXiv)

While imitation learning is becoming common practice in robotics, this approach often suffers from data mismatch and compounding errors. DAgger is an iterative algorithm that addresses these issues by continually aggregating training data from both the expert and novice policies, but does not consider the impact of safety. We present a probabilistic extension to DAgger, which uses the distribution over actions provided by the novice policy, for a given observation. Our method, which we call DropoutDAgger, uses dropout to train the novice as a Bayesian neural network that provides insight to its confidence. Using the distribution over the novice's actions, we estimate a probabilistic measure of safety with respect to the expert action, tuned to balance exploration and exploitation. The utility of this approach is evaluated on the MuJoCo HalfCheetah and in a simple driving experiment, demonstrating improved performance and safety compared to other DAgger variants and classic imitation learning.