CaMeL: Learning Method Preconditions for HTN Planning

Abstract

A great challenge in using any planning system to solve real-world problems is the difficulty of acquiring the domain knowledge that the system will need. We present a way to address part of this problem, in the context of Hierarchical Task Network (HTN) planning, by having the planning system incrementally learn conditions for HTN methods under expert supervision. We present a general formal framework for learning HTN methods, and a supervised learning algorithm, named CaMeL, based on this formalism. We present theoretical results about CaMeL's soundness, completeness, and convergence properties. We also report experimental results about its speed of convergence under different conditions. The experimental results suggest that CaMeL has the potential to be useful in real-world applications.

Robot Controllers for Highly Dynamic Environments with Real-time Constraints

“…In (Ilghami et al 2002;Ilghami et al 2005) the authors propose a learning system for HTNs, where a domain expert solves task nets giving examples to the learner. The learner now generalizes based on the training examples from the human expert and can solve similar tasks in a better way.…”

Learning approximate preconditions for methods in hierarchical plans

“…Ilghami et al (2002) describe an algorithm called CaMeL, which learns preconditions of HTN methods from training data. While CaMeL can in theory achieve 100% precision, it requires a large number of training samples to do so.…”

“…Candidate Elimination requires significant extension for use in an HTN planning context to handle issues like what the representational bias of the version spaces should be, or how the version spaces that represent preconditions of methods in different layers of the task hierarchy should interact with each other. Due to lack of space, we do not describe these extensions, and instead refer interested readers to (Ilghami et al, 2002). …”

“…As shown in (Ilghami et al, 2002), CaMeL sometimes needs a high number of plan traces to achieve full convergence. In this section, we show that the CaMeL++ algorithm requires fewer training examples than CaMeL to solve a similar percentage of planning problems in the test set.…”

“…To overcome this problem, we used the same approach that was used in (Ilghami et al, 2002): To simulate a human expert. We used a correct hierarchical planner to generate planning traces for random planning problems on an HTN domain.…”