Causal Action Influence Aware Counterfactual Data Augmentation

TL;DR

This paper introduces CAIAC, a data augmentation technique that enhances offline robot learning by generating synthetic data through causal influence-based counterfactual reasoning, improving robustness against distributional shifts.

Contribution

The paper presents a novel causal influence-aware counterfactual data augmentation method for offline learning, enabling synthetic data generation without online interactions.

Findings

Increases robustness of offline algorithms against distributional shift

Effective in generating feasible synthetic transitions

Improves generalization beyond training data

Abstract

Offline data are both valuable and practical resources for teaching robots complex behaviors. Ideally, learning agents should not be constrained by the scarcity of available demonstrations, but rather generalize beyond the training distribution. However, the complexity of real-world scenarios typically requires huge amounts of data to prevent neural network policies from picking up on spurious correlations and learning non-causal relationships. We propose CAIAC, a data augmentation method that can create feasible synthetic transitions from a fixed dataset without having access to online environment interactions. By utilizing principled methods for quantifying causal influence, we are able to perform counterfactual reasoning by swapping $\it{action}$-unaffected parts of the state-space between independent trajectories in the dataset. We empirically show that this leads to a substantial increase in robustness of offline learning algorithms against distributional shift.