DropoutDAgger: A Bayesian Approach to Safe Imitation Learning
This work addresses safety concerns in imitation learning for robotics, though it is incremental as it builds on the existing DAgger framework.
The paper tackles the problem of safety in imitation learning by introducing DropoutDAgger, a Bayesian extension to DAgger that uses dropout to estimate action confidence and balance exploration with exploitation, resulting in improved performance and safety in MuJoCo HalfCheetah and a driving experiment compared to other methods.
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.