Jointly Learning Predicates and Actions Enables Zero-Shot Skill Composition
The paper introduces Predicate Action Skills (PACTS), a new approach to learning complex robot behaviors. PACTS jointly models action trajectories and symbolic outcomes, enabling better generalization and skill composition without retraining. This method allows for zero-shot composition of learned skills through effective planning and execution monitoring.
- ▪Learning from Demonstration (LfD) often struggles with generalizing to new skill compositions without retraining.
- ▪PACTS models skills as a joint generative process over action and predicate belief trajectories.
- ▪The approach enhances both action generation and predicate classification, facilitating zero-shot skill composition.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Robotics arXiv:2605.20648 (cs) [Submitted on 20 May 2026] Title:Jointly Learning Predicates and Actions Enables Zero-Shot Skill Composition Authors:Benedict Quartey, Sebastian Castro, Eric Rosen, Wil Thomason, George Konidaris, Stefanie Tellex View a PDF of the paper titled Jointly Learning Predicates and Actions Enables Zero-Shot Skill Composition, by Benedict Quartey and 5 other authors View PDF HTML (experimental) Abstract:Learning from Demonstration (LfD) enables robots to learn complex behaviors from expert examples, yet existing approaches often fail to generalize to new compositions of known skills without retraining.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
