Offline Discovery of Interpretable Skills from Multi-Task Trajectories
This work addresses a central challenge in hierarchical imitation learning for robotics, enabling more efficient skill acquisition from offline data, though it appears incremental as it builds on existing paradigms like VQ-VAEs and option-based frameworks.
The paper tackled the problem of discovering reusable skills from multi-task offline robot demonstrations without explicit rewards or subtask annotations, introducing LOKI, a three-stage framework that achieved high success rates on the D4RL Kitchen benchmark and outperformed standard hierarchical imitation learning baselines.
Hierarchical Imitation Learning is a powerful paradigm for acquiring complex robot behaviors from demonstrations. A central challenge, however, lies in discovering reusable skills from long-horizon, multi-task offline data, especially when the data lacks explicit rewards or subtask annotations. In this work, we introduce LOKI, a three-stage end-to-end learning framework designed for offline skill discovery and hierarchical imitation. The framework commences with a two-stage, weakly supervised skill discovery process: Stage one performs coarse, task-aware macro-segmentation by employing an alignment-enforced Vector Quantized VAE guided by weak task labels. Stage two then refines these segments at a micro-level using a self-supervised sequential model, followed by an iterative clustering process to consolidate skill boundaries. The third stage then leverages these precise boundaries to construct a hierarchical policy within an option-based framework-complete with a learned termination condition beta for explicit skill switching. LOKI achieves high success rates on the challenging D4RL Kitchen benchmark and outperforms standard HIL baselines. Furthermore, we demonstrate that the discovered skills are semantically meaningful, aligning with human intuition, and exhibit compositionality by successfully sequencing them to solve a novel, unseen task.