Problem
Research questions and friction points this paper is trying to address.
Generalizing symbolic actions using entity hierarchy
Learning actions from few observations in simulations
Solving unseen tasks with novel entities and behaviors
Learning Type-Generalized Actions for Symbolic Planning
🤖 AI Summary
In symbolic planning, action representations traditionally rely on expert-crafted, domain-specific encodings, resulting in poor generalizability and high transfer costs across domains.
Method: This paper proposes a type-generalized action learning framework grounded in entity-level hierarchical structure and behavioral similarity—marking the first integration of type generalization into symbolic action learning. It synergistically combines offline inductive logic programming with online dynamic generalization to automatically induce reusable action templates from minimal demonstrations.
Contribution/Results: Evaluated in a simulated kitchen environment, our approach enables zero-shot task solving—including novel entities, long-horizon sequences, and anomalous conditions—without manual encoding. Experiments demonstrate substantial improvements in planning success rates and cross-task generalization over conventional hand-coded approaches, validating both scalability and robustness of learned abstractions.
📝 Abstract
Symbolic planning is a powerful technique to solve complex tasks that require long sequences of actions and can equip an intelligent agent with complex behavior. The downside of this approach is the necessity for suitable symbolic representations describing the state of the environment as well as the actions that can change it. Traditionally such representations are carefully hand-designed by experts for distinct problem domains, which limits their transferability to different problems and environment complexities. In this paper, we propose a novel concept to generalize symbolic actions using a given entity hierarchy and observed similar behavior. In a simulated grid-based kitchen environment, we show that type-generalized actions can be learned from few observations and generalize to novel situations. Incorporating an additional on-the-fly generalization mechanism during planning, unseen task combinations, involving longer sequences, novel entities and unexpected environment behavior, can be solved.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Generalize symbolic actions via entity hierarchy
Learn type-generalized actions from few observations
On-the-fly generalization during planning
🔎 Similar Papers
No similar papers found.
