Quantifying Generalisation in Imitation Learning

2025/09/29 by Gavenski, Nathan, Rodrigues, Odinaldo
#Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2509.24784

Abstract

Imitation learning benchmarks often lack sufficient variation between training and evaluation, limiting meaningful generalisation assessment. We introduce Labyrinth, a benchmarking environment designed to test generalisation with precise control over structure, start and goal positions, and task complexity. It enables verifiably distinct training, evaluation, and test settings. Labyrinth provides a discrete, fully observable state space and known optimal actions, supporting interpretability and fine-grained evaluation. Its flexible setup allows targeted testing of generalisation factors and includes variants like partial observability, key-and-door tasks, and ice-floor hazards. By enabling controlled, reproducible experiments, Labyrinth advances the evaluation of generalisation in imitation learning and provides a valuable tool for developing more robust agents.

Citations

• Meta-World+: An Improved, Standardized, RL Benchmark
• Explorative Imitation Learning: A Path Signature Approach for Continuous Environments
• Imitation Learning Datasets: A Toolkit For Creating Datasets, Training Agents and Benchmarking
• DiffTORI: Differentiable Trajectory Optimization for Deep Reinforcement and Imitation Learning
• Is Imitation All You Need? Generalized Decision-Making with Dual-Phase Training
• Minigrid & Miniworld: Modular & Customizable Reinforcement Learning Environments for Goal-Oriented Tasks
• MAHALO: Unifying Offline Reinforcement Learning and Imitation Learning from Observations
• Benchmarking the Spectrum of Agent Capabilities
• Off-Policy Imitation Learning from Observations
• Imitating Unknown Policies via Exploration
• Leveraging Procedural Generation to Benchmark Reinforcement Learning
• Playing hard exploration games by watching YouTube
• Behavioral Cloning from Observation
• DeepMind Control Suite
• Generative Adversarial Imitation Learning
• Deep Residual Learning for Image Recognition
• High-Dimensional Continuous Control Using Generalized Advantage Estimation
• Playing Atari with Deep Reinforcement Learning
• A Reduction of Imitation Learning and Structured Prediction to No-Regret\n Online Learning
• Neuronlike adaptive elements that can solve difficult learning control problems
• A Survey of Imitation Learning Methods, Environments and Metrics

Related

• The MAGICAL Benchmark for Robust Imitation
• Imitation Learning: Progress, Taxonomies and Challenges
• Generalization Guarantees for Imitation Learning
• Generative predecessor models for sample-efficient imitation learning
• Imitation Learning in the Deep Learning Era: A Novel Taxonomy and Recent Advances