A Large Recurrent Action Model: xLSTM enables Fast Inference for Robotics Tasks

TL;DR

This paper introduces a large recurrent action model with xLSTM that achieves fast, linear-time inference suitable for real-time robotics tasks, outperforming Transformer-based models in speed and comparable in performance.

Contribution

The paper proposes a novel large recurrent action model using xLSTM, enabling fast inference and sequence extrapolation for reinforcement learning in robotics.

Findings

LRAM achieves linear-time inference complexity.

LRAM performs comparably to Transformers on 432 tasks.

LRAM demonstrates faster inference suitable for real-time applications.

Abstract

In recent years, there has been a trend in the field of Reinforcement Learning (RL) towards large action models trained offline on large-scale datasets via sequence modeling. Existing models are primarily based on the Transformer architecture, which result in powerful agents. However, due to slow inference times, Transformer-based approaches are impractical for real-time applications, such as robotics. Recently, modern recurrent architectures, such as xLSTM and Mamba, have been proposed that exhibit parallelization benefits during training similar to the Transformer architecture while offering fast inference. In this work, we study the aptitude of these modern recurrent architectures for large action models. Consequently, we propose a Large Recurrent Action Model (LRAM) with an xLSTM at its core that comes with linear-time inference complexity and natural sequence length extrapolation abilities. Experiments on 432 tasks from 6 domains show that LRAM compares favorably to Transformers in terms of performance and speed.