Q-Learning with VLA Perception for Real-World Navigation Using Low-Cost Sensors

DOI

10.2991/978-94-6239-791-0_3How to use a DOI?

Keywords

Vision-Language-Action Models; Reinforcement Learning; Edge Robotics; Low-Cost Navigation; Latency-Aware Control; Hybrid Architecture

Abstract

Vision-Language-Action (VLA) models allow robots to understand scenes alongside linguistic instructions. These models have computational requirements that are too demanding to implement on edge hardware like a Raspberry Pi and result in an unacceptable latency gap between the interpretation and decision-making stages. Despite GPU acceleration or remote inference, the latency problem still persists, creating delays between perception and action that make closed-loop control systems unstable and unsafe. This paper provides an implementation of a Slow-Fast Hybrid Control Architecture in order to be able to deploy foundation models on low-cost edge devices with remote inference. The slow system running on the cloud uses a highly quantized VLA model that detects objects and gives navigation instructions. The fast system running on the Raspberry Pi uses a Reinforcement Learning policy to provide motor commands in real time, and the RL agent is rewarded using a multi-objective reward function guided by the VLA instructions.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Madhuri Manohar Barhate
AU  - Aditya K. Vasipalli
AU  - Anushri Arora
AU  - Kaivalya Beohar
AU  - Param Bande
AU  - Aaysha Khan
PY  - 2026
DA  - 2026/10/09
TI  - Q-Learning with VLA Perception for Real-World Navigation Using Low-Cost Sensors
BT  - Proceedings of the International Conference on Artificial Intelligence and Machine Learning for Agro-Ecosystems (ICAIML-AE 2026)
PB  - Atlantis Press
SP  - 14
EP  - 24
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-791-0_3
DO  - 10.2991/978-94-6239-791-0_3
ID  - Barhate2026
ER  -