Track: Planning Under a Different Name

AI planning is concerned with using the ability to anticipate the results of actions to solve sequential decision-making problems. ICAPS has traditionally developed representations, algorithms, and theory for this purpose, including classical planning formalisms, heuristic search, planning under uncertainty, and many others. At the same time, over the last decade, closely related planning and deliberation methods have emerged across reinforcement learning, natural language processing, robotics, generative models, program synthesis, and other areas of AI.

Examples of such instances of “planning under a different name” include the combination of learned policies and Monte Carlo tree search, learning world models with deep neural networks, planning in the token space for natural language generation, planning or search over diffusion steps for image generation, chain-of-thought and other forms of test-time reasoning, and many others. Such techniques have led to breakthroughs in fields such as program synthesis, mathematics, and robotics.

ICAPS seeks to strengthen the connection between these developments and the planning community, and to better understand, compare, and unify the underlying ideas. The purpose of the Planning Under a Different Name track is therefore to serve as a bridge between the existing ICAPS community and researchers doing closely related work in other AI communities.

Which Track is Right for Me?

Ideally, all papers submitted to the Planning Under a Different Name track would also be suitable for the main track at ICAPS. However, despite substantial intellectual overlap, the ICAPS community and neighboring AI communities have developed different terminology, research traditions, assumptions, benchmarks, and evaluation practices. As a result, work that is novel and significant in one community may not always be evaluated in the appropriate context by another, and reviewers and authors may end up “talking past” each other during the review process. To address this, reviewers for the Planning Under a Different Name track will come from both the ICAPS community and the relevant neighboring AI communities. They will be instructed to apply the same standards of rigor, novelty, and significance as for other ICAPS submissions, while evaluating contributions in the context of the relevant literature and high-level planning concepts rather than requiring the conventions typically associated with ICAPS papers.

Use the following criteria to determine the right track for your paper:

Suitability for the main track:

• The paper is primarily framed using planning terminology (e.g., PDDL, STRIPS), literature, representations, benchmarks, and evaluation methodology familiar to the ICAPS community.
• Its main contribution can naturally be evaluated using the expertise of the existing ICAPS reviewer community.

Suitability for the Planning Under a Different Name track:

• The work makes a substantive contribution to planning or deliberative sequential decision-making, for example, through search, lookahead, reasoning over alternative trajectories, or predictive models used for decision-making.
• It may involve search over states, actions, tokens, reasoning steps, programs, or latent representations; learned policies, value functions, or heuristics integrated with search; or learned world models used for planning.
• The work is primarily situated in the terminology, literature, benchmarks, or methodological conventions of another research community.
• It may evaluate on domains uncommon at ICAPS, such as language-model reasoning, reinforcement learning, robotics, program synthesis, theorem proving, generative modeling, or other application-specific benchmarks.

The use of machine learning, foundation models, generative models, or sequential computation alone is not sufficient; the work should make a substantive contribution to planning, search, lookahead, deliberation, or to representations, models, or other components intended to enable or improve them.