Neural Algorithmic Reasoners are Implicit Planners

TL;DR

This paper introduces XLVINs, a novel neural planning method that performs value iteration in a latent space, improving data efficiency and aligning with classical planning algorithms in various environments.

Contribution

We propose XLVINs, which perform implicit planning in a high-dimensional latent space, overcoming limitations of prior approaches and enhancing data efficiency in reinforcement learning tasks.

Findings

XLVINs outperform prior implicit planners in low-data settings.

XLVINs closely align with traditional value iteration.

The method improves data efficiency across classical control, navigation, and Atari environments.

Abstract

Implicit planning has emerged as an elegant technique for combining learned models of the world with end-to-end model-free reinforcement learning. We study the class of implicit planners inspired by value iteration, an algorithm that is guaranteed to yield perfect policies in fully-specified tabular environments. We find that prior approaches either assume that the environment is provided in such a tabular form -- which is highly restrictive -- or infer "local neighbourhoods" of states to run value iteration over -- for which we discover an algorithmic bottleneck effect. This effect is caused by explicitly running the planning algorithm based on scalar predictions in every state, which can be harmful to data efficiency if such scalars are improperly predicted. We propose eXecuted Latent Value Iteration Networks (XLVINs), which alleviate the above limitations. Our method performs all planning computations in a high-dimensional latent space, breaking the algorithmic bottleneck. It maintains alignment with value iteration by carefully leveraging neural graph-algorithmic reasoning and contrastive self-supervised learning. Across eight low-data settings -- including classical control, navigation and Atari -- XLVINs provide significant improvements to data efficiency against value iteration-based implicit planners, as well as relevant model-free baselines. Lastly, we empirically verify that XLVINs can closely align with value iteration.