paper
Difference-Aware Retrieval Policies for Imitation Learning
paperactiveprovisional
difference-aware-retrieval-policies-for-imitation-learning-b0ce8e29·1 events·first seen 8d agoAliases: Difference-Aware Retrieval Policies for Imitation Learning
Co-occurring entities
More like this (12)
Imitation LearningPreference-Aware Rubric LearningOne-Shot Imitation LearningReasoning ImitationLearning to Reason by Analogy via Retrieval-Augmented Reinforcement Fine-TuningSelf-Augmenting Retrieval for Diffusion Language ModelsUnderstanding the Behaviors of Environment-aware Information RetrievalRecalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented ModelsReference-Augmented TrainingSparse Subspace-to-Expert Sharing for Task-Agnostic Continual LearningAn Agency-Transferring Model-Free Policy Enhancement TechniqueGeometric Action Model for Robot Policy Learning
Recent events (1)
DARP: Semi-parametric retrieval-based imitation learning reduces compounding errors by 15-46%
Researchers introduce DARP (Difference-Aware Retrieval Policies), a semi-parametric imitation learning method that retrieves k-nearest neighbor demonstrations at inference time and predicts actions based on relative distance vectors between neighbor and query states. The approach reparameterizes behavior cloning around local neighborhood structure rather than global state-to-action mappings, requiring no additional data collection or online expert feedback. Across continuous control and robotic manipulation tasks, DARP shows 15-46% performance improvements over standard behavior cloning, including on high-dimensional visual inputs.