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

2024-10-29

arXiv.org

AI Summary

Transformer-based architectures suffer from high inference latency, hindering their deployment in real-time robotic closed-loop control. Method: This work pioneers the integration of xLSTM into large-scale action modeling for robotics, replacing self-attention with xLSTM to achieve linear-time inference complexity and robust long-horizon extrapolation, while leveraging an offline reinforcement learning framework to ensure both training scalability and real-time online inference. Contribution/Results: Evaluated on a comprehensive benchmark comprising 432 robot control tasks across six domains, the proposed model matches Transformer performance while substantially accelerating inference—reducing end-to-end latency to the millisecond level. This marks the first practical deployment of a large action model for real-robot closed-loop control, bridging the gap between foundation-model capabilities and real-time robotic autonomy.

Technology Category

• Intelligent Robots: Behavior Learning & Control
• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

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.

Problem

• Addresses slow inference in Transformer-based RL models
• Proposes xLSTM for fast real-time robotics applications
• Demonstrates LRAM outperforms Transformers in speed and performance

Innovation

• xLSTM for fast inference
• Linear-time complexity model
• Natural sequence extrapolation